A Comprehensive Digital Mirror of Reality

The intersection of chemistry and artificial intelligence (AI) is a fascinating area that attracts a lot of attention in both research and industry. We talked to people working in the field about the potential of AI to revolutionize chemical research, but also concerns, (current) limitations, and ethical implications for chemical applications. We also asked for ideas to try or experiment with, as well as useful articles and videos for beginners and advanced users.

Professor Markus Kraft, University of Cambridge, UK, works in the fields of knowledge graphs, combustion, chemical engineering, artificial intelligence, and automated laboratories.

What fascinates you about AI?

AI requires an understanding of the world, which I always found fascinating, and it is probably the ultimate motivation for our curiosity. One intriguing conceptualization that has significantly influenced my perspective is Marvin Minsky’s book “The Society of Mind“. Marvin Minsky articulates AI as a collection of smaller processes, or agents, where each agent individually performs simple tasks that need no consciousness at all. Yet, joining these agents in specific configurations yields genuine intelligence.

In my endeavor, The World Avatar project, my colleagues and I have been working toward the realization of artificial intelligence through the orchestrated assembly of autonomous agents. This approach is analogous to orchestrating a symphony, where each agent plays a unique role. Our preliminary research in this direction has been promising, and the prospect of achieving an all-encompassing digital replica of the real world is genuinely thrilling.

Combining these ideas with powerful machine learning algorithms, such as generative AI, which have recently become prominent, will accelerate this process. Witnessing the convergence of these techniques into a coherent intelligence mirrors the convergence of diverse elements in the real world. I am very excited to see how this research unfolds in the next ten years.

Is there anything we should fear?

As AI systems grow more competent and have the ability to alter the world, including human society, it is important that the AI’s goals are in line with the goals of a thriving humanity. The renowned “Three Laws of Robotics” proposed by Isaac Asimov provide a conceptual foundation, yet the practical implementation of these principles in AI systems remains a challenge. Particularly in light of the United Nations’ Sustainable Development Goals (SDGs), bridging the gap between the qualitative articulation of our objectives and their quantitative interpretation by robots and AI systems poses an ongoing challenge.

The initiation of global dialogues on AI regulations by governments worldwide, as evidenced in global AI summits, signifies a commendable first step. However, we must proceed with caution in the trajectory of developing AI systems, especially those with the potential for artificial general intelligence.

Our group has made initial steps in quantifying some of the SDGs metrics and monitoring them within The World Avatar, but this is not an easy task. The evolving nature of the real world necessitates continuous scrutiny to ensure that its trajectory is faithfully reflected in the cyberspace, so that the decisions made by the agents are well-informed and align with human values and societal well-being. Striking a delicate balance between innovation and ethical governance is paramount as we navigate the uncharted terrain of AI’s evolution.

Do you have something for our readers to try out or experiment with?

Certainly! I invite your readers to explore our latest iteration of the Marie website, a cutting-edge question-answering system tailored for chemistry enthusiasts. This website empowers users to ask questions in natural language. Behind the scenes, our language model translates these questions into a query language tailored for our knowledge graph. The system then retrieves precise answers, mitigating the risk of “hallucinations” often associated with language models.

What sets Marie apart is its commitment to transparency—every piece of information retrieved is accompanied by a clearly established provenance. This ensures a trustworthy and verifiable source for the answers provided.

Currently, we are actively expanding the system’s capabilities to tackle more intricate queries that leverage cross-domain information. This represents a significant leap forward, addressing challenges that traditional technologies struggle to handle. I encourage your readers to have a try with this system and stay tuned for the exciting developments on the horizon.

Can you recommend good articles for beginners and one you enjoyed recently?

For beginners entering the field of AI in chemistry, I recommend exploring the following papers:

From my group:

From Platform to Knowledge Graph: Evolution of Laboratory Automation (perspective paper):
This paper reviews community efforts in developing self-driving laboratories, analyzing the limitations of the current platform-centric approach and proposing the potential of dynamic knowledge graph technology in achieving a globally collaborative research network.
- From Platform to Knowledge Graph: Evolution of Laboratory Automation,
  Jiaru Bai, Liwei Cao, Sebastian Mosbach, Jethro Akroyd, Alexei A. Lapkin, Markus Kraft,
  JACS Au 2022, 2, 292–309.
  https://doi.org/10.1021/jacsau.1c00438
Knowledge Engineering in Chemistry: From Expert Systems to Agents of Creation (perspective paper):
This paper introduces the history and the core principles of knowledge engineering, as well as its applications within the broad realm of chemical research and engineering. It outlooks how encoding wisdom of the human experts in structuring knowledge can transform the way we conduct research.
- Knowledge Engineering in Chemistry: From Expert Systems to Agents of Creation,
  Aleksandar Kondinski, Jiaru Bai, Sebastian Mosbach, Jethro Akroyd, Markus Kraft,
  Acc. Chem. Res. 2022, 56, 128–139.
  https://doi.org/10.1021/acs.accounts.2c00617
From Platform to Knowledge Graph: Distributed Self-Driving Laboratories (research article):
This paper develops an architecture to support distributed self-driving laboratories. It demonstrated a use case where two robots, located more than 10,000 km apart on the globe, collaboratively optimized a chemical reaction upon a goal request from scientists while adhering to the FAIR data principles.
- From Platform to Knowledge Graph: Distributed Self-Driving Laboratories,
  Markus Kraft, Jiaru Bai, Sebastian Mosbach, Connor Taylor, Dogancan Karan, Kok Foong Lee, Simon Rihm, Jethro Akroyd, Alexei Lapkin,
  Res. Sq. 2023.
  https://doi.org/10.21203/rs.3.rs-3141873/v1

I also particularly enjoyed reading these papers from the community:

Chemputation and the Standardization of Chemical Informatics (perspective paper):
This paper connects computation and chemical automation, leading to “chemputation”. It advocates for a publication standard to exchange chemical code for synthesis to ensure reproducibility and interoperability.
- Chemputation and the Standardization of Chemical Informatics,
  Alexander J. S. Hammer, Artem I. Leonov, Nicola L. Bell, Leroy Cronin,
  JACS Au 2021, 1, 1572–1587.
  https://doi.org/10.1021/jacsau.1c00303
Integrating autonomy into automated research platforms (perspective paper):
This paper discusses the challenges in bridging the gap between automation and autonomy in laboratory research. It highlights the need for flexible and goal-oriented chemical programming to achieve and accelerate autonomous research in novel domains.
- Integrating autonomy into automated research platforms,
  Richard B. Canty, Brent A. Koscher, Matthew A. McDonald, Klavs F. Jensen,
  Digit. Discov. 2023, 2, 1259–1268.
  https://doi.org/10.1039/D3DD00135K
The rise of self-driving labs in chemical and materials sciences (review paper):
This paper provides a roadmap for implementing self-driving laboratories for non-expert scientists. This review discusses the status quo of successful implementations, their current limitations, and future opportunities in the field.
- The rise of self-driving labs in chemical and materials sciences,
  Milad Abolhasani, Eugenia Kumacheva,
  Nat. Synth. 2023, 2, 483–492.
  https://doi.org/10.1038/s44160-022-00231-0
In Pursuit of the Exceptional: Research Directions for Machine Learning in Chemical and Materials Science (perspective paper):
This paper discusses the integration of machine learning, robotic experimentation, and human knowledge for fully automated discovery workflows. This paper conjectures the potential role of large language models in facilitating the planning and orchestrating chemical researches.
- In Pursuit of the Exceptional: Research Directions for Machine Learning in Chemical and Materials Science,
  Joshua Schrier, Alexander J. Norquist, Tonio Buonassisi, Jakoah Brgoch,
  J. Am. Chem. Soc. 2023, 145, 21699–21716.
  https://doi.org/10.1021/jacs.3c04783

These resources provide a diverse perspective on the current challenges and innovative approaches within the AI in chemistry domain. I hope they inspire and guide beginners’ exploration in this exciting field.

Is there anything else you would like to share with readers of ChemistryViews?

Yes, I encourage readers to explore our recent preprint titled “The Digital Lab Framework as part of The World Avatar”, where we delve into the rationale behind advocating for a holistic integration of all aspects in a scientific laboratory as a crucial element in shaping a sustainable future for humanity. Our focus extends to the potential of dynamic knowledge graph technology in realizing this vision.

Specifically, we took a systems engineering perspective, where we conducted a thorough strengths and weaknesses analysis of both existing solutions and our dynamic knowledge graph approach. A key aspect of The World Avatar approach involves aligning with overarching goals and establishing an infrastructure that enables the decomposition of abstract goals into manageable tasks with tangible outcomes.

We firmly believe that our approach fosters connectivity with other initiatives in the community, creating a synergistic effect that amplifies collective efforts. In this spirit, I extend an invitation for collaboration within the community. We can build an interoperable system together that facilitates the collective contributions of diverse stakeholders. I believe this will propel us towards a future where advancements in AI and lab automation converge for the betterment of scientific exploration and innovation.