Ever wonder whether computers will one day be capable of doing everything that human beings can? If so, pick up the recent book by engineer and computer scientist Dr. Robert J. Marks: “Non-Computable You: What You Do That Artificial Intelligence Never Will.”
Marks explains what makes human beings unique, and therefore why no computer will ever match all human capabilities. To be sure, computers excel humans at many tasks — but only tasks that are “algorithmic,” or that entail step-by-step instructions to complete, such as calculating probabilities, retrieving information, or executing functions.
That is why advanced artificial intelligence (AI) systems have long surpassed humans in games such as Chess and Go. These games have more possible moves than there are atoms in the universe, and calculation optimization is crucial for success.
Nevertheless, many capacities of the human being will forever remain beyond the power of AI, capacities like understanding, sentience, creativity, emotion, and common sense. Why? Because these things are “non-algorithmic,” meaning they are not computable.
Take understanding, for example. A computer can follow a step-by-step set of instructions (an algorithm), but it cannot grasp the meaning of those instructions.
To understand why, consider an algorithm that consists of a series of zeros and ones. The only reason such a series counts as an algorithm in the first place is that some human being programmed the digits to act as one. But by itself, the sequence is meaningless.
For example, “0” is a conventional symbol we have invented to convey the meaning of “zero.” But without such a convention, without us inputting the meaning of “zero” to “0,” the symbol is an arbitrary squiggle devoid of semantic content. Of course, algorithms are comprised of such symbols, which means they cannot operate apart from programmers who imbue such sequences of digits with meaning.
All these point to this conclusion: While computers can carry out algorithms, there is no understanding of the meaning of the algorithm within the algorithm itself. Rather, its meaning resides in the mind of the programmer. As Oren Etzioni, the director of the Allen Institute for AI, notes, artificial intelligence “recognizes objects, but can’t explain what it sees. It can’t read a textbook and understand the questions in the back of the book.”
Marks helps the reader see this in a different way by noting the trouble AI has in disambiguating the meaning of unclear sentences, including botched news headlines:
A fun example of AI’s lack of common sense when faced with ambiguity is flubbed headlines. Seattle’s Microsoft, Amazon, and Boeing are headquartered near the coast in the state of Washington. Yet Seattle businesses were not concerned when faced with the headline ‘Tuna Biting Off Washington Coast.’ Why? Because of course tuna were not chomping off big chunks of Seattle beaches. We use common sense to identify the intended meaning — good news for fishermen! — and the incorrect interpretation makes us smile. But AI can’t recognize ambiguity. It lacks common sense.
Because AI is incapable of understanding, programming AIs to contextualize words is no easy task. This explains why computers tend to have trouble with puns, jokes, riddles, and sarcasm. Of course, it is possible and indeed is increasingly common for advanced algorithms to put words into the right context to properly interpret ambiguities. But this only proves AI programs are improving at mimicking understanding — not that AI is in fact growing in understanding.
Furthermore, an especially relevant part of the book challenges another popular idea: that AI is “creative.” For example, it is commonly believed today that AI creates art, music, and even screenplays. But if creativity involves thinking outside the box, then AI fails the creativity test, or the “Lovelace test.”
That asks whether an AI does something that is beyond either the intent or the explanation of the programmer. For instance, certain AI programs have produced what seem like genuinely novel pieces of artwork, so one may be tempted to conclude they can act creatively. Yet, according to the Lovelace test, real creativity requires that an AI act outside of its programming.
Take AlphaGo, an advanced AI developed to master the game of Go. If AlphaGo were capable of genuine creativity, it would not merely beat the best humans at Go but act independently of its programming altogether. Marks writes that if AlphaGo were to, say, “beat all comers at the simple game of Parcheesi, the Lovelace test would be passed. But as it stands, AlphaGo is not creative. It can only perform the task it was trained for, namely playing GO. If asked, AlphaGo is unable to even explain the rules of GO.”
On the other hand, humans regularly act creatively according to this metric. While every person is capable of creativity, Marks mentions obvious examples to highlight the point. Pioneers like Archimedes, Nikola Tesla, Albert Einstein, and Bob Dylan (to name a few) challenged the conventional wisdom of their time by stepping “outside the box” of consensus to revolutionize their fields. AI, however, will never do such a thing because it remains bound by the parameters of its programming.
Consciousness, emotion, and wisdom are additional dimensions of the human person that will forever evade even the most advanced AI systems. These things, like understanding and creativity, are non-algorithmic, meaning they cannot be programmed.
But aside from describing what makes humans unique, the book also introduces the reader to the basics of computational theory and the history of artificial intelligence. For those not technologically inclined, these sections may have less appeal. Nevertheless, Marks does a nice job of dumbing down the technical stuff so it is both readable and interesting for the non-specialist.
Most importantly, however, “Non-Computable You” disabuses us of the mistaken belief that human beings are simply complex computers, and that AI will one day be capable of doing everything we can — only better. The truth is that there are dimensions to the human person that no AI, no matter how advanced, will ever be able to replicate.
David formerly worked at a public policy institution. Follow him on Twitter @DWeinberger03. Email him at davidweinberger916@gmail.com.
Comments are closed.