Autonomous experimental systems in materials science

ABSTRACT The emergence of autonomous experimental systems integrating machine learning and robots is ushering in a paradigm shift in materials science. Using computer algorithms and robots to decide and perform all experimental steps, these systems require no human intervention. A current direction focuses on discovering unexpected materials and theories with unconventional research approaches. This article reviews the latest achievements and discusses the impact of autonomous experimental systems, which will fundamentally change the way we understand research. Moreover, as autonomous experimental systems continue to develop, the need to think about the role of human researchers becomes more pressing. While machine learning and robotics can free us from the repetitive aspects of research, we need to understand the strengths and limitations of machine learning and robots and focus on how humans can perform higher creativity. In addition, we also discuss inventorship and authorship in the era of autonomous systems. GRAPHICAL ABSTRACT


Introduction
The total number of all possible materials is estimated to be at least 10 60 [1,2].This number illustrates the vastness of the materials search space, which must contain many materials that can help address current societal problems.In a way, the world of materials is a frontier for exploration, much like space or the deep sea.
How can we quickly and systematically find unexpected materials within this enormous search space?To this end, materials science needs a tool that can transcend the limits of human capabilities to serve as a materials explorer (Figure 1), akin to a spaceship or a deep-sea exploration vessel.
The core of the materials explorer is the autonomous experimental system based on machine learning and robots (green, orange, and purple in Figure 1).Here, the term 'autonomous' means that a computer algorithm decides the next experimental steps, and robots perform all experimental steps.This approach that involves no human intervention is called the closed-loop experiment (Figure 2).
In general, new materials are searched in a multidimensional space by optimizing many relevant experimental parameters.Because of the vastness of the search space, the manual optimization of these parameters by individual researchers only produces incremental results that do not show the big picture.However, this problem is ideally suited for the autonomous experimental system to address. Figure 2 illustrates one such example.Here, based on the initial instructions, machine learning decides which compound to synthesize and feeds the corresponding directions to the robots; the robots synthesize, test, and report the results back to the algorithmrepeating the cycle until the desired result is obtained.This autonomous experimental approach drastically speeds up the materials exploration processes.
The materials explorer will fundamentally change the way we understand and conduct research in the following three stages.
Stage 1: Optimization of the yield of target substances Here, the target compound is known, but the optimum synthesis conditions are unknown.The target compound is decided by the human researchers before the experiment, and the autonomous experimental system quickly optimizes the synthesis conditions of the target compound within the search space specified by the human researchers [3].
Stage 2: Finding new materials with desired properties Here, the target physical properties are decided, but the compound possessing these properties is unknown.The autonomous experimental system quickly finds the best material within the search space specified by the human researchers.In contrast to Stage 1, it is the composition of the materials that is changed to find new compounds to meet the required physical properties.Materials with a variety of crystal structures and hierarchical structures are also explored.
Stage 3: Finding new materials or principles that no one has thought of before The vision of the materials explorer.The exploration involves an autonomous experimental system, materials informatics, and human researchers.The heart of the materials explorer is an autonomous experimental system based on machine learning and robots (green, orange, and blue).This system is imbued with the skills of experts and generates large amounts of experimental data that could not have been generated by human researchers (data-production factory).The data generated by the autonomous experimental system is then processed by machine learning and simulations to predict new materials (materials informatics).In addition, the system organizes the data and generates 'materials maps' and models, facilitating knowledge creation by providing researchers a sharable big-picture view of unexpected materials, thereby accelerating materials development.

Figure 2.
The concept of autonomous experimental system.The system autonomously synthesizes materials with optimal physical characteristics without human intervention.Autonomy leads to significant improvements: 1) fully digitalized experiments transforms all experimental parameters, including process conditions, into data; 2) removal of human error makes reproducibility reliable; and 3) implicit knowledge can be digitalized and embedded.
Here, new materials, theories, and principles that are unexpected to researchers are discovered by combining the results from autonomous experiments, materials informatics, and human researchers (Figure 1) [4].
Stage 3 is rapidly advancing owing to the development of machine learning, robotics, and materials informatics.In this stage, it is critical to embed the researchers' intuition and experience into the construction of theories.This inclusion of the human role is often referred to as the 'human-in-the-loop' or the 'researcher-in-the-loop'.
An important point to note is that the transformation resulting from the three stages will change the way researchers think, giving completely new perspectives that researchers cannot obtain using the conventional research method.In addition, the transformation is not limited to a single laboratory; instead, it will change how we conduct research through a materials digital transformation.For example, experimentalists can remotely fabricate materials via the internet.In the same way, theorists can fabricate materials to test their predictions.
The data generated by this system is combined into big data.Researchers use this big database to extract humanunderstandable information (materials informatics).Because machine learning and robotics alone can neither find insights nor discover concepts in physics and chemistry, human researchers will always remain central to the research.The key point is that materials scientists must understand what machine learning and robotics can solve, and must set the right problem to be solved.The strength of human researchers lies in concept creation or problem identification in the larger context.Combining these strengths with machine learning and robotics is critical to accelerating research (researcher-in-the-loop).
The autonomous materials exploration raises one fundamental question; when experiments are performed autonomously and new materials are found without human intervention, who is the discoverer and inventor?This question not only relates to the authorship of papers and the inventorship of patents, but it also concerns the researchers' motivations.It is important to address this discussion as the autonomous experimental approach develops.
In this article, we review the trends and the prospects of autonomous experimental systems in materials science, specifically in following aspects: • The overall status of autonomous experiments in materials science.• The history and the latest topics of autonomous materials synthesis.• The future of research using the autonomous experimental system.
• Some important take-home lessons we have learned when developing and using such a technology.• The future of authorship and inventorship.
Throughout this review, we address what human researchers should focus on in the era of autonomous research.The year 2020 was the year of game change; there were significant advancements in the field of autonomous research.This review aims to contribute to the opening of the era of autonomous materials research.

A brief description of our own autonomous experiments
In the closed-loop cycle shown in Figure 2, researchers only need to choose the material properties to optimize and provide the system with the necessary raw materials; the automatic system then takes control, repeatedly synthesizing and measuring the properties of new compounds until the best one is found.The machine-learning algorithm uses previous knowledge to decide how the synthesis conditions should be changed to approach the desired outcome with each cycle.
Recently, we have developed an autonomous synthesis of inorganic thin films [3].In this proof-of-concept study, we demonstrated the autonomous fabrication of TiO 2 thin films with low resistance and showed that this system accelerates experiments by tenfold.
To obtain these results, we have used robotic modules of a sputter deposition apparatus and a robotic device for measuring resistance.Other modules with robotic synthesis and measurement equipment can be connected to this system to adapt to the desired research.The robotic arm transfers the samples from module to module as needed, and the Bayesian optimization algorithm predicts the synthesis parameters for the next iteration.

Stage 1: Optimization of the yield of target substances or reaction conditions
The idea of autonomous materials synthesis using robots has a long history.The year 1978 saw the first proposal on a fully automated closed-loop robot aiming for the optimization of chemical reaction parameters [31].However, no experimental results were reported.
In the 1980s, Matsuda et al. reported the optimization of reaction conditions using an autonomous system [18].The repetition of reagent adjustment, reaction, measurement, and prediction of the next experimental conditions based on the measured results were automatically performed.Here, test tubes were manipulated by a robot.The color-developing reactions for chemical analysis were optimized using the simplex method with three parameters: the amount of two reagents and the reaction time.The exhaustive grid search required 130 experiments, but the robot system optimized the reaction in less than 28 experiments.While the purpose of this experiment was not to find a new compound but to optimize the color-developing reaction, it is a seminal pioneering work of autonomous experiments.
In 2016, Maruyama et al. demonstrated an autonomous synthesis of inorganic materials.The authors synthesized carbon nanotubes using chemical vapor deposition [19].They fabricated multiple columns containing a catalyst layer on a substrate in advance.Then, they heated the individual columns one by one with a laser while repeatedly moving the substrate to synthesize carbon nanotubes with different growth conditions.This heating laser was also used as an excitation source for Raman spectroscopy to observe the growth rate in situ.A genetic algorithm maximized the growth rate; the system optimized the temperature, pressure, and gas composition.Later the group utilized a Bayesian optimization as an optimization algorithm [32].
For organics, Hein et al. maximized the yield of the Suzuki-Miyaura coupling reaction using Bayesian optimization [20].They used a commercially available liquid handling robot and ChemOS [33][34][35] for autonomous experiments.In the Bayesian optimization, Phoenics [36] and Gryffin [37] algorithms were used to optimize categorical variables (catalyst type) in addition to continuous variables (amount of catalyst, amount of feedstock, reaction temperature).Parallel autonomous process optimization experiments in batches were performed to shorten the time to complete the optimization.
The autonomous synthesis of organics using flow reactors was reported by Jensen et al. [21] The authors first reported on a Heck reaction, where the yield was maximized by optimizing the raw materials ratio and reaction time as independent variables, using the simplex method (Nelder-Mead method).Subsequently, the authors worked on other autonomous syntheses using flow reactors [38,39].For example, they applied SNOBFIT (Stable Noisy Optimization by Branch and Fit) to a variety of reactions using a modular system [40].In 2015, Cronin et al. optimized the yield of an imine synthesis [22].The authors also used the simplex method to tune the raw materials ratio and reaction time.
The system was equipped with in-line nuclear magnetic Optimizing the amount of reagent, reaction time, etc. using the simplex method Maruyama (2016) [19] Laser to perform both heating and spectroscopy Carbon nanotubes with maximized growth rate (Stage 1) Optimizing temperature, pressure, and gas composition using a genetic algorithm Hein (2021) [20] Liquid handling robot Maximizing the yield of Suzuki-Miyaura coupling reaction (Stage 1) Optimizing continuous variables (amount of ingredients, etc.) and categorical variables (catalyst type) using Bayesian optimization Jensen (2010) [21] Flow reactor (Synthesis in liquid phase) Maximizing the yield of Heck reaction (Stage 1) Optimizing temperature and raw materials ratio using simplex method Cronin (2015) [22] Flow reactor Maximizing the yield of imine synthesis (Stage 1) Optimizing temperature and raw materials ratio using simplex method Clare, King (2009) [23,24] Liquid handling robot, a robotic arm, etc.
Discovering yeast genes (Stage 2) Generating hypotheses and experimentally testing these hypotheses Cooper (2020) [25] Free-roaming robot Photocatalyst mixtures with maximized photocatalytic activity (Stage 2) Optimizing the concentration of photocatalyst and additives using Bayesian optimization Berlinguette (2020) [26] A gripper, a pipette mount, and a robotic arm Organic hole transport materials with maximized hole mobility (Stage 2) Optimizing annealing times and dopant concentrations using Bayesian optimization deMello (2007) [27] Flow reactor CdSe nanoparticles with targeted spectroscopic characteristics (Stage 2) Optimizing temperature and precursor concentrations using SNOBFIT method Kumacheva (2021) [28] Flow reactor Au nanoparticles with targeted spectroscopic characteristics (Stage 2) Optimizing reaction time and precursor concentrations using Bayesian optimization Wright (2013) [29] Flow reactor Abl kinase inhibitors with maximum activity (Stage 2) Synthesizing from 27 × 10 row materials using random forest Viswanathan (2020) [30] Flow reactor Aqueous electrolytes with maximum electrochemical window (Stage 2) Optimizing each precursor solution volume using Bayesian optimization Hitosugi, Shimizu A robot arm, a sputtering system, a resistance meter Nb-doped TiO 2 thin film with minimized resistance (Stage 2) Optimizing oxygen partial pressure using Bayesian optimization resonance.The group also developed an organic synthesis robot that navigates chemical reaction spaces [41][42][43].

Stage 2: Finding new materials with desired properties
In 2009, the Clare and King group reported a seminal work [23,24], where a robot 'Adam' autonomously generated functional genomics hypotheses and experimentally tested the hypotheses using robots.The system measured the growth curves of selected microbial strains growing in defined media, 'discovering' three novel yeast genes.The equipment comprised a liquid handling robot, a robot arm, and an incubator -all fixed in one place.
In 2020, Cooper et al. demonstrated a free-roaming robot that moved around the laboratory to perform autonomous experiments using the same equipment as those used by its human counterparts [25].The robot aimed to maximize photocatalytic activity by optimizing the concentrations of photocatalyst and additives.Based on Bayesian optimization, the robot identified the photocatalyst mixtures that were six times more active than the initial formulation.The robot completed 688 experiments in 8 days.This number of experiments would take a human researcher several months.For reagent weighing, the robot handled both powder and liquid materials.
In the same year, Berlinguette et al. reported the autonomous fabrication of organic thin films [26].A robot 'Ada', equipped with a gripper, a pipette mount, and a robotic arm for liquid injection and substrate transfer, maximized the hole mobility of organic hole transport materials used in perovskite solar cells.The dopant concentrations and annealing time were optimized using Bayesian optimization.Ada finished the experiment in five days instead of nine months.Later, the group used Ada to define a Pareto front of conductivities and processing temperatures for palladium films formed by combustion synthesis [44].There are other liquid handling robots developed for closedloop, e.g.Aspuru-Guzik et al. reported an automated platform for organic laser discovery, including not only synthesis and compound identification but also integrated target property characterization [45].
For organic compounds, Wright et al. synthesized Abl kinase inhibitors with maximum activity in 2013 [29].Out of 270 possible combinations (10 types × 27 types) from the raw materials, the system found the novel Abl kinase inhibitor after synthesizing only 21 compounds.The prediction model used random forest regression to handle the types of raw materials with different structures.
Viswanathan et al. reported isolating aqueous electrolytes with the maximum electrochemical window in 2020 [30].The authors used Bayesian optimization to optimize the solution volume of three aqueous Li salts or four aqueous Na salts.The results indicated that nonsmooth chemical responses are observed along the axis of the amount of NaBr.The authors pointed out that logscaling, which varies rapidly with quantity, such as the amount of NaBr, gives the response surface both smaller gradients along this axis and a much-improved performance because Gaussian process regression requires an assumption of smoothness on the response surface.They also pointed out that the selection and presentation of the design space for materials search are of utmost importance to autonomous task design.

Future of Stages 1 and 2: Autonomously search within the preset search space
Currently, Stages 1 and 2 are not widespread because the range of applicable experiments is still limited.As more sophisticated robots are developed, the application of autonomous research will expand in turn.The problem lies in that conventional robots are rigidly designed for precision, aiming for zero error.Because they are programmed based on if-then statements, these robots cannot respond flexibly to changing situations.Therefore, it is difficult to implement the tacit knowledge of skilled human researchers in robot systems.With soft control, soft structure, multiple sensing devices, and machine learning, a more flexible robot can absorb errors and disruptions, and can operate without stoppage or breakdowns [51,52].Thus, developing robot technology that can respond flexibly until the goal is achieved is important.To this end, it is essential to develop new techniques for robots to learn the minute details of any required motions.
When versatile robots capable of performing a variety of experiments become affordable, the introduction of robots will rapidly expand into many research fields.At the time of this writing, the cost of robots is reducing rapidly to the point where labs can consider using such robot-experiment automation.However, more effort is needed to help materials scientists implement robots into their experiments.
We note that not all researchers will adopt this style of research.The conventional research approach, which relies on researchers' experience, knowledge, and intuition, will remain essential for conceiving new synthetic methods and analytical techniques.We expect there will be a hybrid between conventional and autonomous approaches.
Autonomous experimental systems are not only useful for performing high-throughput experiments; the most critical aim is to help human researchers deepen their research.Human researchers need to think about how to analyze the collected data, how the new materials can be used, and what theories can be deduced from unexpected results.Autonomous experimental systems free human researchers from repetitive tasks and enable researchers to conduct innovative research.Furthermore, combining the system with advanced materials prediction based on computer simulation is the step that enables human researchers to perform more creative work (Figure 3).
With the implementation of autonomous experimental systems, theorists can test their original ideas.Once theorists predict a promising material, they can log on to the autonomous experimental systems via the internet and synthesize the new material.Currently, we are developing a system capable of automatically synthesizing materials from simple instructions, such as 'synthesize A x B y C z with a crystal structure of D'.The system automatically measures X-ray diffraction patterns and feeds the results back to the algorithm to predict the next synthesis condition.

Stage 3: Finding materials that no one has thought of before
ecause the autonomous experimental systems will generate increasingly larger amounts of data, it is essential to combine this type of system with materials informatics and computer simulations to extend human thinking.In so doing, the generated data will promote the emergence of unexpected results that are beyond conventional theories and experiences.It is also important to embed the researchers' intuition and experience, which are tacit knowledge, into the autonomous system and share them with other researchers (Figure 4).The following are some of the remaining challenges.A vision of how research will be conducted.In the conventional way of research, there has been a disruption between researchers and robots/AI (artificial intelligence).In the near future, researchers develop materials based on an enormous amount of experimental and simulation results.In this researcher-in-the-loop process.Human researchers will obtain new insights, a bird's eye view of materials, though multifaceted materials characterization and resulting materials big data.

Increasing the amount of reliable materials data (materials checkup system and measurement)
To date, materials informatics is still limited by the lack of a reliable materials experimental database.Such a database cannot be manually acquired.
To this end, autonomous experimental systems enable the storage of all data, ranging from the fabrication processes (synthesis conditions and experimental environment) to the structural and composition information, to the measured materials properties.Even the negative data (data that do not show the desired values) are stored for later use.This inclusive approach to data storage drastically increases the amount of data available for accurate prediction based on machine learning.In addition, all data are highly reproducible because they are not acquired by humans (i.e.devoid of human errors).Therefore, it is possible to analyze in detail which process parameters are correlated with physical properties, leading to technology for predicting synthetic processes (process informatics).
The sheer increase in experimental speed directly translates to an increase in the amount of materials data.The speed of autonomous experiments is estimated to be at least one order of magnitude faster than conventional experiments; it is possible to accelerate the acquisition of measurement data by another order of magnitude using the materials checkup system (Figure 5) [3].
In conventional research, human researchers often need to transport their samples to multiple analysis systems to evaluate the physical properties.To avoid errors associated with sample transport and shorten the time for analysis, all analytical modules should be connected to each other.Connecting multiple materials property measurement devices and automatically measuring one after another enable multifaceted materials characterization (Figure 5).With the recent development of measurement and analysis informatics [53], it is possible to perform measurements with even higher accuracy, precision, and sensitivity measurements in a short time [54].
The key to the multifaceted evaluation is to increase the probability of finding unexpected results.We have already experienced finding an unexpected electrode material using multifaceted evaluation while targeting the synthesis of a solid electrolyte.

Constructing scientific theories from a bird's eye view
The autonomous experimental system does not search aimlessly in the search space.Instead, it creates a materials map [55] based on the large amount of described in the previous section to give a bird's eye view of the materials world (Figure 6).
Specifically, the materials informatics organizes the enormous amount of multi-dimensional data (actual experimental data and high-throughput first-principles simulations) into human-understandable forms.This data is overlaid with previously reported experimental and simulated data to create a materials map.
The materials map depicts the untapped and promising areas that should be experimentally verified.This depiction stimulates the researchers' thinking in ways conventional research methods cannot.The new ideas can then be immediately tested using the autonomous experimental system (Figure 1).
The unveiling of the hidden correlations and relationships between materials properties and processes will promote the construction of new theories.Until now, completely unexpected and never-observed-before results are often ignored because it has been difficult to distinguish whether they are real or whether they are  caused by experimental error.However, these results can be reproduced when using the autonomous experimental system; unexpected results (i.e.'outliers') will not be ignored, making it possible to create new scientific theories that explain them (Figure 7).Lastly, to overview a massive amount of materials data, it is necessary to have the technology capable of converting this data for humans to understand.In particular, the development of technologies for explainable machine learning and visualization of information in a multi-dimensional space through dimensionality reduction is mandatory.

Important points for the actual operation of an autonomous system
This section discusses some important take-home lessons we have learned from our constructing and operating the autonomous experimental system.

Setting the right research topic
It is not practical to leave all experimental operations to machine learning and robots; instead, it is crucial to choose the right tasks.The current autonomous experimental technology works best when applied to processes with well-developed instructions.Techniques with broad and general purposes many researchers use are better suited for applying autonomous systems.For example, we chose to automate sputtering thin-film deposition because it is widely used in the field, including the semiconductor industry.
Another point concerns the human operator's psychological aspect.In our experiments, the human operators perform two tasks: setting up the raw materials and cleaning up the pieces used in the experiments.Because the operator performs these tasks according to the robot's schedule, in some cases, the operator may feel as if being controlled by robots.It is critical that robots perform these tasks in the future.

Cost issues
It is difficult to introduce an autonomous system to every laboratory because of its cost.There are three points to consider.The first point is that robots are becoming less expensive, and their range of applications is expanding.The price depends on positioning accuracy and repeatability.Because robots work 24 hours a day, 7 days a week, the operational cost depends on the tasks and the labor cost.The software cost has also decreased, as robots can be run using LabVIEW or Python -two widely-used programming environments.In this way, the total system cost is becoming affordable, and we are now in an era when robots can be introduced to laboratories.We note that it is important to provide open labs for researchers unaware of the possibilities of autonomous systems, to get acquainted with the actual setups and operations of such systems.
The second point is to promote sharing the autonomous system via the internet to provide access to the system from anywhere in the world (cloud laboratory or experiments over the cloud, Figure 8).Currently, the equipment in the laboratory is only used during the daytime on weekdays and therefore is not fully utilized.Full utilization around the clock through sharing will enhance cost-effectiveness; it will also enable materials fabrication while researchers are at home, making remote work style possible.
The last point is the perspective of multifaceted characterization.As shown in Figure 5, multiple analytic instruments automatically characterize a range of properties.The time previously spent by researchers on measurements can now be allocated to higher value-added tasks -further enhancing cost-effectiveness.with some amount of data, hypotheses can be proposed.Typical researchers could not predict that the number '1' is hidden in this figure .Only an exceptional researcher can recognize the hidden '1'.Materials science to date has been 'local (belong to particular researchers)' in this way.(c) When big data is available, researchers can easily recognize '1', without relying on such exceptional researchers.Furthermore, by considering the 'outliers (shown in red)', we can see the true situation -a '4' rather than a '1'.We can thus construct new theories (big-picture materials science).

Change in the experimental sequences
Conventionally, the materials to be synthesized are decided before the synthesis.After the synthesis, compositions and structures are analyzed to confirm that the aimed materials have been synthesized.After identifying the compounds, the properties are evaluated.
Autonomous experiments reverse this sequence.Because the goal of the research is to determine the material with superior physical properties, there is no need to decide what materials to synthesize in advance.The compositional and structural analyses to identify the materials are only performed once the robot has synthesized materials with the optimum properties.
Autonomous experiments applied to device fabrications show a similar sequence reversal.The system automatically fabricates a device, evaluates its performance, and tunes the next materials based on the results.Again, there is no need to know the details of the materials prior to device fabrications.
From a broader perspective, autonomous systems change our attitude toward research.New ideas -that previously required careful experimental planning to be tested with a small number of trials -can now be tested immediately using an autonomous system.With automation, we can comfortably conduct experiments even with a naive idea because the autonomous system performs experiments, and the number of experiments can be increased at will.

Steps for implementation and the rise of the lab-system integrator
How should we start the automation and autonomous experiments?The following points should be considered when integrating materials synthesis and evaluation, robotics, machine learning, and overall system control.
(1) Clarify the workflow and choose the appropriate tasks for automation (see section 'Setting the right research topic').( 2) Set up a team including an engineer who understands informatics.In addition, include an engineer familiar with equipment control and systemization.(3) The introduction of robots can be expensive.It may be difficult to introduce the whole system at once.Therefore, it is necessary to create a chain of successes to maintain continuous funding.The critical point is to draw an overall plan and accumulate small successes.
Standardizing the lab equipment is essential to reduce the time to set up and the cost.Shortly, labsystem integrators will emerge.There is a movement in the world towards aligning the format of all mechanical interfaces and communication protocol; each piece of equipment will be integrated to form an automation system [15].Standardization and de facto standards are key for lab-system integrators, and the technology must be launched quickly.

The autonomous system and intellectual property
Currently, the intellectual property of newly discovered or invented materials is effectuated through scientific publication or patent application.However, when the discovery/invention is accomplished without human intervention, can we say that a human discovered or invented the materials?Can the autonomous experimental system be an inventor under patent law [25]?Can the system be an author of a paper?Furthermore, patents are designed to protect inventions that are not only new, but also involve an inventive step; will the autonomous system change how Present Future inventive steps are determined?-inother words, will the autonomous system change the standard of what constitutes an easy invention?In this section, we discuss inventorship, inventive step, and authorship.

Inventorship -Who is an inventor?
Whether artificial intelligence including an autonomous system (hereafter referred to as AI) should be treated as an inventor under patent law is a hot topic [56][57][58].A person who makes an invention is called the inventor (we note that the patent owner is often not the inventor but the company that employs the inventor).However, if AI makes the invention, can it be considered the inventor under patent law?Abbott et al. applied for patents naming AI 'DABUS' as the inventor in several countries and regions to raise this issue [59].They proposed that the inventor is DABUS and the patent owner would be the person who owns DABUS.For this case, the South African patent authorities have accepted the patent applications with DABUS as the sole inventor [60]; we note that the decision can still be overturned by the respective superior courts.
Since the wording of patent acts and their interpretations differ from one jurisdiction to another, whether AI can be treated by patent acts as an inventor also differs.In contrast to the above examples, the patent offices and courts in the U.S., U.K., Australia, the European Patent Office, and the Japan Patent Office have all ruled that only human beings can be the inventor under the patent acts, rejecting the notion of DABUS being treated as an inventor [58,59,[61][62][63].In such countries/regions, even when a new material is discovered by an autonomous experimental system, only human researchers who have operated the system and contributed creatively may be treated as inventors.As mentioned in the section 'Autonomous experimental systems in the world', the current level of technology still needs significant human contribution to conduct the research.
However, there is an argument that if the autonomous experimental system becomes more autonomous and the human contribution continues to decrease, the human researcher may no longer be treated as the inventor.For example, in Japan, the inventor is recognized as a person who conceives an invention or creates the specific materials based on the conceived idea [64,65].From this perspective, if the autonomous experimental system both conceives and creates a specific material, the human would no longer be considered the inventor.As the autonomous system evolves, in order to be recognized as the inventor, future human researchers will have to either exercise creativity independently from the autonomous experimental system, or demonstrate that the autonomous experimental system is used as a tool.

Inventive step -Is an invention easily invented?
To be protected by a patent, an invention must satisfy the definition of inventive step (non-obviousness in the US or Canada).The question here is in what way the autonomous experimental system will change how the inventive step is determined.In the following, we discuss the impact of the system on the inventive step in the future, according to the Japanese patent system.
By definition, an inventive step requires that an ordinary skilled person in the field not be able to make the invention easily at the time of filing the patent application.In Japan [66], patent examiners determine the inventive step by assessing (1) factors indicating the absence of inventive step, and (2) factors supporting the existence of inventive step.Specifically, factor (1) questions whether a skilled person in the field could have easily made the invention, and factor (2) questions, for example, the existence of advantageous effects exceeding predictability based on the state of the art.
At present, because AI in materials exploration is still in its infancy and its accuracy remains low, the current patent examination practices determining inventive step have not been changed.However, many researchers in intellectual properties and business associations point out that, as AI's accuracy continues to rise and becomes widely used, the difficulty of achieving new invention will be lowered for the skilled person in the field -raising the standard for what satisfies the definition of inventive step [67][68][69][70][71].In this case, how the inventive step is legally determined may be changed in the future.
Here we apply this argument to the finding of novel materials by the autonomous experimental system.At this time, the system is not yet widely used, but, in the future, the situation may change as this system becomes well implemented or shared in laboratories.Then, we believe the determination of the inventive step may be changed by a shift in the balance between factors (1) and (2).For example, because the autonomous experimental system excels at optimizing numerical conditions, it is conceivable that many invention cases involving such optimizations would be accessible to an ordinary skilled person in the field with access to the autonomous experimental system.Such cases would make factor (1) stronger because it is sufficiently reasoned for the ordinary skilled person to arrive at the invention, causing the standard for satisfying the inventive step to be raised.Although the possibility to surpass the raised standard by considering a combination of factor (2), such as advantageous effects, the chance of satisfying the inventive step decreases if the current research and development approach remains unchanged.
To obtain patent protection in a future where the autonomous experimental system is widely used, human researchers will have to achieve the inventive step by redirecting their creativity away from the strengths of the autonomous experimental system -in addition to the inventorship mentioned in the section 'Inventorship' above.For example, the autonomous experimental system will never be able to conceive highly innovative synthesis techniques, or develop new characterization techniques.Furthermore, even for inventions involving optimization of numerical conditions, human creativity will contribute to the inventive step when it is difficult even for the autonomous experimental system to come up with a specific parameter.In addition, new materials discovered by autonomous experimental system will still need to be put to use, requiring human input to create new devices by combining many materials, utilizing hierarchical structures.Lastly, it is human beings who will think about what issues facing society and how to solve them; thus, the role of the human researchers is to judge the value and to decide what to do next.This big-picture perspective will lead to new inventions that only humans can think of.

Authorship -Who is an author of academic papers?
Each journal and publisher has its authorship guidelines.One well-known example is the authorship guidelines published by Committee on Publication Ethics (COPE) [72], which state that 'the minimum requirements for authorship, common to all definitions, are (A) substantial contribution to the work and (B) accountability for the work that was done and its presentation in a publication'.Another well-known example from International Committee of Medical Journal Editors (ICMJE) [73] also includes the same concept as requirements (A) and (B).
Although most scientists disapprove of articles crediting the AI tool as an author, an AI tool such as ChatGPT has listed its name as an author [74-76].However, these cases do not focus on the autonomous experimental system but mainly show that AI can generate sentences.Whether or not the autonomous experimental system can be treated as an author of an academic discovery is a quite important issue and should be discussed.
So, let's consider the case where human researchers write a manuscript on a new material discovered by the autonomous experimental system.At the current technological level, while the autonomous experimental system can satisfy the substantial contribution (requirement A) because the system has performed the laboratory procedures leading to the new-material discovery, it cannot provide accountability (requirement B).
There are many theories on whether AI can have accountability [77]; however, at present, the general view is that only human researchers can be held accountable [78].Until AI can provide accountability, AI is not eligible for academic authorship.

Summary
As the involvement of machine learning and robotics continues to expand in scientific research, the roles of human researchers have become a central question.Machine learning and robots will not replace researchers; from the onset, autonomous experimental systems have aimed at providing researchers with the chance to think and to be more creative.The important point is that scientists understand both the limitations and the possibilities of machine learning and robotics, and apply these techniques to the right problem.Even in an age where robots and machine learning perform most of the laboratory procedures, human researchers will always be the main players.This approach to automation and autonomy will exponentially accelerate creative research and development, contributing to the future development of science, technology, and industry.

Figure 1 .
Figure 1.The vision of the materials explorer.The exploration involves an autonomous experimental system, materials informatics, and human researchers.The heart of the materials explorer is an autonomous experimental system based on machine learning and robots (green, orange, and blue).This system is imbued with the skills of experts and generates large amounts of experimental data that could not have been generated by human researchers (data-production factory).The data generated by the autonomous experimental system is then processed by machine learning and simulations to predict new materials (materials informatics).In addition, the system organizes the data and generates 'materials maps' and models, facilitating knowledge creation by providing researchers a sharable big-picture view of unexpected materials, thereby accelerating materials development.

Figure 3 .
Figure3.The human researchers will always play the leading role.The autonomous laboratory is for researchers.This technology makes it possible to narrow down candidate materials with high predictive performance and to pinpoint materials synthesis experiments.The researchers then work on highly creative tasks and acquires 'tacit knowledge' of an even higher level.In turn, the new knowledge is digitized again.It is essential to repeat this cycle.

Figure 4 .
Figure 4.A vision of how research will be conducted.In the conventional way of research, there has been a disruption between researchers and robots/AI (artificial intelligence).In the near future, researchers develop materials based on an enormous amount of experimental and simulation results.In this researcher-in-the-loop process.Human researchers will obtain new insights, a bird's eye view of materials, though multifaceted materials characterization and resulting materials big data.

Figure 5 .
Figure 5. Concept of the materials checkup system for finding unexpected results.All the material-property measurements are done automatically [3], copyright 2020 Ryota Shimizu et al. and reprinted with permission under a Creative Commons Attribution (CC BY) license.

Figure 6 .
Figure 6.Creation of a 'materials map' to identify unexplored materials.Autonomous experiments enable the creation of materials maps using experimental data.We put emphasis on unexpected experimental results and explore their surroundings.

Figure 7 .
Figure 7. Importance of outliers.Suppose there is a number hidden in the tiles in (a).With no data, nothing can be predicted.(b)with some amount of data, hypotheses can be proposed.Typical researchers could not predict that the number '1' is hidden in this figure.Only an exceptional researcher can recognize the hidden '1'.Materials science to date has been 'local (belong to particular researchers)' in this way.(c) When big data is available, researchers can easily recognize '1', without relying on such exceptional researchers.Furthermore, by considering the 'outliers (shown in red)', we can see the true situation -a '4' rather than a '1'.We can thus construct new theories (big-picture materials science).

Figure 8 .
Figure 8. Experiments in the cloud space.Sharing the autonomous system via the internet has become possible.

Ryota
Shimizu is an associated professor at School of Materials and Chemical Technology, Tokyo Institute of Technology.He received Ph.D (Doctor of Science) from The University of Tokyo in 2011.His research fields include solid-state physics and chemistry at surfaces and interfaces based on thin film technology, and materials informatics.Taro Hitosugi is a professor at the Department of Chemistry, The University of Tokyo.He also serves as a Specially Appointed Professor at the School of Materials and Chemical Technology, Tokyo Institute of Technology.He received his Ph.D. from The University of Tokyo in 1999 and joined Sony Corporation.He specializes in solid state chemistry and interface science.He has published more than 170 refereed papers in academic journals.

Table 1 .
Examples of autonomous materials synthesis using robots.