AI Is a Test for Humanity. Will It Pass?

Published on Industrial Biotechnology (Joseph Byrum)

Alan Turing created his famous test for artificial intelligence nearly 70 years ago [1]. It was a simple method, not for determining whether a computer program could actually think, but whether the machine was smart enough to pass for a human. Computer scientists are still trying to beat this challenge [2].

It’s only a matter of time before machine learning exploits ongoing advances in processing power to surpass this benchmark. And when that happens, roles will be reversed as the very existence of a fully functional artificial intelligence becomes a test for humanity. The idea of a machine that could pass for a human was nothing more than a thought experiment in the 1950s, but it’s far more than that now.

We’ve come a long way in just the past decade, with everything up to, and including, the kitchen sink [3] integrating with virtual assistants like Alexa. Soaring sales figures suggest consumers view these products as good enough at parsing human speech and providing somewhat relevant answers, even though their robotic voices and occasionally bizarre answers aren’t likely to fool anyone into thinking they’re human. In fact, when I asked Alexa, whether or not ‘‘she’’ could pass the Turing Test, she responded that she didn’t need to, since she’s ‘‘not pretending to be human.’’

Researchers, on the other hand, keep trying to pretend. An MIT deep learning algorithm can create handwriting [4] and sound that are good enough to fool humans, though the systems still fall short when it comes to a believable text conversation [5]. Given such progress, it’s not hard to foresee in our  lifetimes an AI that can pass for human; not by having the sophisticated tricks of an advanced chatbot, but by using language to communicate in a genuine way. That is, the machine would have something to say, and it would intuitively know how to say it.

The concept of ‘‘intuitive AI’’ is catching on in popular culture. While AI has historically assumed the role of villain in blockbuster films, a kinder, more human form is represented on our screens. For example, the NBC series, The Good Place, takes a philosophical look at what it means to be a good person, with the afterlife as a backdrop. In the show’s second season, actor Ted Danson’s virtual assistant, Janet, after a series of reboots and malfunctions (long story), learns to express emotion. It becomes a more human version of her AI-self, communicating authentically as the season progresses.

If a similar scenario does happen in the future, society must come to terms with a digital device that can do what humans do best— think—and do it better than we can. Surely that would be as disturbing to us as it was for many in the 16th century to discover that the earth is not at the center of the universe, or waking up in the 18^th to find that the Industrial Revolution has made our job obsolete.

We’ve even seen similar events in biotechnology. For centuries, the breeding of plants has been a labor-intensive process conducted by expert human breeders. They relied upon their intuition to decide which plants to crossbreed to create offspring with the greatest chance of having desired genetic traits. Only recently has this time-tested, old-fashioned process been uprooted by the creation of tools that rely on data analytics to eliminate the guesswork entirely. The choices of what plants to breed, when to breed them, by what method, and so on could be made by software tools that produce demonstrably more efficient results than a human operating alone [6].

The response from those human breeders to this development? Despite some initial resistance to the idea of machines moving in on their territory, in the labs where this system is used, the human breeders have realized the tool makes them more effective. They still have a role to play—pollination is still a manual process—and their skills are put to good use completing a complex process with an even higher success rate.

So, it’s not necessarily inevitable that we will see machine-smashing and riots when robots start taking away even those high-end creative jobs in which to succeed, the machine would have to pass for a human.

It’s also unlikely that we will reach the point in which an AI system could communicate genuine insight. To build such a system, we would first have to gain a deeper understanding of knowledge and replicate it in a machine. In other words, for AI to advance to this next level, we must advance in our understanding of ourselves. One would hope with this greater understanding, humanity would pass its test—as it has done before.

Learning that the earth was not the center of the universe kicked off the Copernican Revolution and inspired incredible scientific achievement. Society likewise survived the economic displacement of the Industrial Revolution to enjoy an unprecedented rise in the world’s standard of living.

So, while it’s tough to think we might one day face digital competition in the creativity department, there’s good reason this digital competitor will help drive progress by augmenting human abilities, rather than supplanting them. As long as society keeps an open mind to the potential for progress, humanity will have a chance of passing this test.

REFERENCES

[1] Could a computer think? Available at: http://turing.org.uk/scrapbook/test.html (Last accessed September 2019).

[2] The Society for the Study of Artificial Intelligence and Simulation of Behaviour (2018). Available at: https://aisb.org.uk/new_site/?page_id=2 (Last accessed September 2019).

[3] Clark Thompson A. Delta to make faucets that work with Amazon Alexa (2018). Available at: https://cnet.com/news/delta-to-make-faucets-that-work-withamazon-alexa/ (Last accessed September 2019).

[4] Hardesty L. Computer system passes ‘‘visual Turing test’’ (2015). Available at: http://news.mit.edu/2015/computer-system-passes-visual-turing-test-1210 (Last accessed September 2019).

[5] Knight W. Tougher Turing Test Exposes Chatbots’ Stupidity (2016). Available at: https://technologyreview.com/s/601897/tougher-turing-test-exposeschatbots-stupidity/ (Last accessed September 2019).

[6] Byrum J, Davis C, Doonan G, et al. Advanced Analytics for Agricultural Product Development (2016). Available at: ttps://pubsonline.informs.org/doi/abs/10.1287/inte.2015.0823 (Last accessed September 2019).