Article: Should We Put a Pause on AI Research?

Photo Source: Unsplash

On March 22, 2023, The Future of Life Institute published a letter titled “Pause Giant AI Experiments: An Open Letter.” The Future of Life, a think tank founded in 2014 to protect the world from dangerous technological advancements, published the letter with over 1,200 signatures, including those of Elon Musk, the founder of Tesla and SpaceX, Steve Wozniak, Co-founder of Apple, and Yoshua Bengio, a pioneer in artificial intelligence. (futureoflife.org) The letter calls for a six-month moratorium on training larger and larger AI models. The letter cites earlier work done with the AI research community to draft principles to develop AI for the betterment of humanity.

The letter states, “AI systems with human-competitive intelligence can pose profound risks to society and humanity, as shown by extensive research and acknowledged by top AI labs. As stated in the widely-endorsed Asilomar AI Principles, Advanced AI could represent a profound change in the history of life on Earth, and should be planned for and managed with commensurate care and resources.”

The letter implies that the rapid development of large language models poses a threat. It states, “Contemporary AI systems are now becoming human-competitive at general tasks, and we must ask ourselves: Should we let machines flood our information channels with propaganda and untruth? Should we automate away all the jobs, including the fulfilling ones? Should we develop nonhuman minds that might eventually outnumber, outsmart, obsolete, and replace us? Should we risk loss of control of our civilization? Such decisions must not be delegated to unelected tech leaders.”

The staggering artificial intelligence arms race flooding the news in recent months feels almost out of the blue. Following the mercurial rise of interest in OpenAI’s chatGPT since its launch to the public in November of 2022, millions of users have signed up to use this large language model. The big tech players such as Meta, Google, and Microsoft have spent billions of dollars in research acquisitions to build larger and larger AI systems. These systems have produced remarkable results such as human conversational capabilities, essay composition, computer code creation, and writing in the style of different authors. Such an unprecedented technological expansion has prompted the call for a six-month moratorium on model development.

In science and technology, such a moratorium is not unique. In 1973, Herbert Boyer of the University of California and Nobel Laureate Stanley Cohen of Stanford University ushered in the age of genetic engineering by successfully transferring DNA from one organism to another in a process called cloning. Cloning fundamentally changed biology, medicine, and agriculture by allowing man to manipulate the blueprint of life. Such potential in the hands of a few scientists caused shock and alarm. The pioneers of genetic engineering keenly felt the potential for their new technology to cause harm. The public also feared the unintended consequences of altering the environment with genetically modified plants, engineering more virulent diseases for germ warfare, and changing evolution in unpredictable ways. To address these fears, Boyer, Cohen, and other leading scientists of the day organized a symposium. The symposium occurred at Asilomar Conference Center near Monterey, California in 1975. It gathered scientists, ethicists, government officials, and journalists to discuss the potential dangers of genetic engineering and to propose actions to protect against these dangers.

Astonishingly, before the Asilomar Symposium, scientists worldwide volunteered to participate in a self-imposed moratorium on genetic engineering experimentation until the conference. The concern was for the safety of the scientists performing the experiments, people, and the environment in general. Asilomar was a ground-breaking event because it was the first time scientists recognized the potential danger of their new technology and organized a public debate to address the issues. A core accomplishment of the symposium was the drafting and agreement on fundamental safety issues, including the proper laboratory procedures for handling potentially dangerous biomaterials, experimental design to prevent the accidental spread of modified organisms, and a moratorium on the cloning of genes from pathogenic organisms such as smallpox. Remarkably, Asilomar lasted three days, but the conclusions and guidelines paved the path for remarkable growth and safety in biotech life sciences and agriculture for the following 50 years.

Since the first Asilomar Symposium on genetic engineering in 1975, science has taken other pauses, such as the ban on using CRISPR to edit germline cells. CRISPR stands for an experimental technique developed from bacterial DNA that allows precise editing of genetic sequences inside living cells. Such a technique opened the door to permanently editing people’s genetic sequences. The ability to modify human genetics to change elements such as eye color, disease risk, and any other heritable attribute of a child caused great concern among the discoverers of CRISPR. In 2015, Chinese scientists attempted to modify non-viable human embryos with CRISPR to reverse a lethal genetic disorder called beta-thalassemia. The embryos did not survive, and the CRISPR research community came together in 2015 at the International Summit on Human Gene Editing to debate the ethics of genetic modification of human embryos. The Summit concluded that CRISPR used to modify embryos is irresponsible both from an evolutionary perspective due to potential unintended consequences and from a social perspective through the concept of designer babies. As a result, the Summit cooled research in embryo genetic modification. However, in 2017 a Chinese researcher did modify the genes of two female embryos, Lula and Nana, to make them resistant to HIV infection. The experiment did not work perfectly and received worldwide condemnation with calls for a complete moratorium on this type of research.

The call for a six-month moratorium on building large language models bigger than GPT4 initiated by the Future of Life Institute cites the imminent danger of such models to cause harm to individuals and society. Such hazards include the convincing spread of misinformation, toxic and hateful speech, propagation of incorrect or misleading information, and usurping creative industries by sucking up artists’ work and repackaging it without credit or remuneration to the creators. Following the open letter from the Future of Life Institute, Time magazine published an opinion piece on March 23, 2023 by Eliezer Yudkowsky, a machine learning researcher. In the essay, Yudkowsky considers a six-month moratorium insufficient to address the safety issues posed by artificial intelligence that he fears at some point will “kill everyone.” (time.com)

A timely six-month pause sounds like a good idea for the industry to consider the ramifications of these increasingly complex and capable models, but not for the general reasons mentioned above. People must remember that these increasingly impressive machines remain machines, and their size and complexity do not hold a candle to the complexity of living organisms. A large language model builds up statistical relationships between words. As the model gets bigger, it generates very realistic responses to questions like solving crossword puzzles, but the model has no real experience. For people, a language composed of words or numbers allows us to summarize and communicate in an imperfect manner our internal and external experiences. When people say, “There are no words to describe how I feel” or “I feel blue today,” they imperfectly communicate their experiences. A large language model can use these exact words in the proper context without knowing the experience. Just like the dizzy, vertigo feeling that many feel standing at the edge of a steep drop like the Grand Canyon will not be felt by the model because it cannot know the fear of falling. A moratorium should give us pause to look at ways that such machines as GPT4 distract us from the truth and cause our ability to think critically to atrophy. Many seem overjoyed to hand off essay writing to a machine because it makes many uncomfortable; however, writing and struggling to form our ideas makes us stronger and more capable of making decisions and expressing ourselves.

Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high-level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.

You can buy his book on Amazon in paperback and in kindle format here.