Worry About Your Average Day, Not Smart AI

Published on Industrial Biotechnology (Joseph Byrum)

The Vocabulary Problem: Distinguishing AI Types

Perhaps the greatest threat to advancing biotechnology through artificial intelligence (AI) is the looseness of our vocabulary. The late physicist Stephen Hawking, one of the greatest minds of our time, [1] once suggested AI ‘‘could be the worst event in the history of our civilization.’’ Not the Black Death, not one of the World Wars, not the Holocaust. A machine that can think for itself.

AGI vs. Today’s AI: Separating Science Fiction from Reality

Hawking was referring to a theoretical possibility involving a very specific form of AI that doesn’t yet exist—artificial general intelligence (AGI). This term is used to describe a machine that can think for itself with human-like self awareness. Hawking reasoned such a device by its nature could ‘‘re-design itself’’ at such an alarming rate that it would become unstoppably powerful.

The Current State of Artificial Intelligence

We’re nowhere near the point where AGI is possible, much less a threat to humanity. The first step in making a machine think like a human is to replicate the way our brain works in hardware and software. Unfortunately, our knowledge of the inner workings of the mind remains limited, [2] and we can’t replicate what we don’t even understand. We’re so far away from AGI that we don’t have an idea how long it would take to create one—if it is even possible to do so at all.

That means AGI isn’t a near-term threat, but fear of the more general term ‘‘artificial intelligence’’ invites real peril right here and now. Popular culture lumps any slightly autonomous system, no matter how rudimentary, under the ‘‘AI’’ banner. Fear of AGI spreads unwarranted distrust, as if a consumer-grade smart toaster or Alexa device is going to rise up against us one day.

Augmented Intelligence: The True Power of Today’s AI Systems

At an industrial scale, augmented intelligence systems, the most powerful form of AI that’s available right now, combine a machine’s memory and computational abilities with the judgment of skilled human operators. The potential for these systems to enhance productivity and positive outcomes is unmatched. In health care, for example, augmented intelligence is being deployed to detect disease in early stages and develop new drugs as remedies.

FDA Adaptation to AI-Driven Medical Innovation

The Food and Drug Administration (FDA) has approved the QuantX AI software to assist radiologists in detecting breast cancer, [3] and it has given the go-ahead to IDx-DR, an AI software package that detects eye problems related to diabetes [4]. The agency earlier this year realized its entire regulatory structure revolves around static medical devices, as opposed to the dynamic solutions that AI offers, so it is the process of updating the rules to better accommodate AI as integral to the future of medicine [5].

AI-Powered Drug Discovery and Development

Pharmaceutical firms like Genentech and GNS Healthcare are using a causal machine learning and simulation AI platform [6] to treat cancer patients. The system automates the process of sifting electronic medical records and genetic data and simulating the effect of different interventions on individual patients to ‘‘reverse engineer’’ the disease. GNS recently raised $23 million in funding for its effort, [7] joining a list of 158 startups who have turned to AI to develop drug treatments [8].

AI is needed to deal with the inherent complexity of biology. The European Bioinformatics Institute, a repository for data sources in the life sciences, contains 40 biomolecular databases that exceed 273 petabytes in total [9]. That’s ten times the size of all the books in the Library of Congress [10]. There’s no way for humans to make use of data on this scale without the assistance of AI.

Navigating Biological Complexity: The Data Challenge

Plant biology is no less complex. The datasets involved in plant genomes, climate modeling, soil quality and crop history are massive. And, just as in the field of medicine, increasing positive outcomes is a matter of life and death. Our current rate of crop production simply isn’t enough to feed everyone as the global population increases by two billion by the year 2050, by the United Nations’ estimate [11].

Agricultural Applications: Feeding a Growing Population

Scaling up production using existing technology and farming methods is easy—if you can throw a few million more acres of land, millions of pounds of surplus fertilizer and millions of extra gallons of fresh water at the problem. Taking all those resources is neither easy nor realistic. Given the real-life constraints on resources, the solution is to make plants grow larger in harsher conditions with fewer inputs while boosting the efficiency of the rest of the production chain.

Plant Breeding Revolution: AI-Driven Genetic Improvements

As with medicine, AI is stepping up to the challenge by creating an unprecedented data harvest for the breeding of superior plant genetics. Algorithms are harnessing massive stores of data and using simulation to identify which plant varieties should be crossbred with one another. Such systems consider every factor, down to the impact of various government regulations, in determining how, when and where testing should take place to see if the proposed plant varieties live up to their potential.

Democratizing Precision Farming Through AI

AI can also lower the barrier to entry for precision farming, giving every farmer the tools needed to bring more product to market in the most efficient and profitable way possible. AI algorithms are ideally suited for absorbing massive amounts of data from every potentially relevant source from which the system can provide customized advice and timely insight for growers regarding which seeds to use, when they should be planted, when they should harvest and so on. Every farmer who has a cell phone essentially can have the world’s top agronomist in his pocket.

Traditionally, farmers think in terms of their field, but the combination of AI, sensors and high-tech machines make it possible to grow on a plant-by-plant basis. A plant on one side of a field might need more water and nutrients than a plant on the other side. AI can take care of making the necessary adjustments, while always measuring performance and fine-tuning its own algorithms. Thus, grower-centric service platforms will manage profitable crop growth at an unprecedented scale.

Challenging Our Assumptions About AI Risk

So, what happens if society, out of an abundance of fear, were to avoid AI solutions in health care and agriculture? After all, many already oppose genetic modification of plants out of a similar fear. Forget about saving lives with early disease detection and treatment. At a minimum, two billion people are going to go hungry.

Starvation from technological stagnation is a bigger threat than HAL 9000.

In fact, we ought to worry more about human miscalculation than the rise of smart machines. In the United States, about 161,000 people die every year from mishaps—like falling out of bed. Such preventable accidents are the third leading cause of death [12] behind heart disease and cancer. Worry more about clumsiness than killer AGI.

The Path Forward: Responsible AI Development in Biotechnology

In fact, the idea that a truly intelligent AI would of necessity want to wipe out the human race is an unsupported assumption. It’s at least as likely that a self-aware computer would become a thinker, augmenting its own knowledge because that is the highest good of a machine, working with humans, or ignoring them at worst.

Still, it’s hard to combat the negative ‘‘approval rating’’ that AI has, courtesy of decades of movies and books in which the technology plays the villain. In the biotech sector, we need to focus on how many lives would be lost unless we continue to develop and deploy augmented intelligence systems.

REFERENCES

[1] Kharpal A. Stephen Hawking says A.I. could be ‘worst event in the history of our civilization’ (2017). Available at: https://cnbc.com/2017/11/06/ stephen-hawking-ai-could-be-worst-event-in-civilization.html (Last accessed November 2019).

[2] Sigfried R. There’s a long way to go in understanding the brain (2017). Available at: https://sciencenews.org/blog/context/neuroscience-understanding-brain (Last accessed November 2019).

[3] US FDA. Device Classification under Section 513(f)(2)(de novo) (2019). Available at: https://accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/denovo.cfm?ID= DEN170022 (Last accessed November 2019).

[4] US FDA. FDA permits marketing of artificial intelligence-based device to detect certain diabetes-related eye problems (2018). Available at: https://fda.gov/ news-events/press-announcements/fda-permits-marketing-artificialintelligence-based-device-detect-certain-diabetes-related-eye (Last accessed November 2019).

[5] FDA. Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) – Discussion Paper and Request for Feedback. Available at: https:// regulations.gov/document?D=FDA-2019-N-1185-0001 (Last accessed November 2019).

[6] GNS Healthcare. GNS Healthcare Announces Collaboration to Power Cancer Drug Development with REFS Causal Machine Learning and Simulation AI Platform (2017). Available at: https://gnshealthcare.com/gns-healthcareannounces-collaboration-to-power-cancer-drug-development/ (Last accessed November 2019).

[7] P&T Community. GNS Healthcare Raises $23 Million Led by Cigna Ventures (2019). Available at: https://ptcommunity.com/wire/gns-healthcare-raises-23- million-led-cigna-ventures (Last accessed November 2019).

[8] Smith S. 167 Startups Using Artificial Intelligence in Drug Discovery (2019). Available at: https://blog.benchsci.com/startups-using-artificial-intelligence-indrug-discovery (Last accessed November 2019).

[9] European Bioinformatics Institute. Annual Report 2018 (2018). Available at: https://ebi.ac.uk/about/digital-bookshelf/publications/EMBLEBI_Scientific_Report-2018.pdf (Last accessed November 2019).

[10] Ashenfelder M. Transferring ‘‘Libraries of Congress’’ of Data (2011). Available at: https://blogs.loc.gov/thesignal/2011/07/transferring-libraries-of-congress-ofdata/ (Last accessed November 2019).

[11] United Nations (Growing at a slower pace, world population is expected to reach 9.7 billion in 2050 and could peak at nearly 11 billion around 2100 (2019). Available at: https://un.org/development/desa/en/news/population/ world-population-prospects-2019.html (Last accessed November 2019).

[12] Centers for Disease Control. Leading Causes of Death. Available at: https:// cdc.gov/nchs/fastats/leading-causes-of-death.htm (Last accessed November 2019).