A recent article contains a great deal of truth, which the French summarize: Plus ça change, plus c'est la même chose: Funding is the constant that drives AI and the choice of terminology.
Note to Alex: When you're defining the terminology, be sure to include the price tags. See below for some excerpts from a recent article. From the founders in 1956 to today, it's la même chose. John ___________________ If You Only Learn One Historical Fact About AI, Let It Be This One: AI’s greatest lie and greatest success, by Alberto Romero: https://albertoromgar.medium.com/if-you-only-learn-one-historical-fact-about-ai-let-it-be-this-one-4373e94a5092 . . As a scientific field, AI (also computer science more generally, let’s ascribe blame where it’s due) has spent its history coining terms that have semantically blurred what happens inside in an attempt to close the gap with the cognitive sciences while going toward a divergent goal: Instead of understanding the human brain through explanatory theories, like neuroscience and psychology, AI is trying to artificially build one without necessarily understanding anything. Some broadly known examples of this semantic similarity are “neural networks” and “machine learning,” popularized a few decades back; “language models,” “attention mechanisms,” and “emergent behavior” have been established more recently. “Hallucination” won’t be the last, but it’s the first one that has created a backdoor that allows us to look directly into the makers’ facade and use it against them, as Klein has aptly done above. Because, in some sense, all these anthropomorphizing concepts are also open windows to hallucinating about a future that may never come. I realized the depth of this trap when I was reading the comments section to check out people’s (contrarian) takes on Klein’s conclusion. I didn’t go too far — The Guardian had pinned a comment that, in its unintended irony, sparked the idea to write this piece. This is the first sentence of that comment: “I have worked in AI and am extremely concerned at the idea that we hand over a high degree of autonomy and power to machines without any semblance of moral or democratic debate about whether we should be doing so.” It’s amazing to me how a reader who clearly agrees with Klein’s thesis implicitly proved the ultimate consequences of her initial point in such a short amount of words: “Hand over … power to machines.” Not to corporations. Not to tech CEOs. Not even to the designers and engineers immediately behind the machines — as if we were already talking about human-level artificial agents to which we can “hand over power.” Klein attacked current AI boosters’ hallucinations about the future, but it’s in the past where the biggest hallucination took place. In anthropomorphizing the field with those brainy terms since its conception, the AI community also hallucinated — as Klein explains with a swift shift of the word’s meaning — a hypothetical future that not even actively critical insiders can adequately decouple from the reality we’re actually heading to. If even those who remain skeptical despite the absurd amount of hype and who ostensibly despise the consequences that Klein highlights are unaware of how much anthropomorphism is ingrained in their conception of AI, there’s not much Klein can do to get her message across. And how could it be otherwise when the single most successful, “powerful and enticing cover stor[y],” as Klein refers to hallucinations, is also the original anthropomorphism — the one that has hopelessly influenced, during 70 years, everything that has come afterward — the greatest trap, the greatest lie: calling AI “AI.” From marketing it was born, in marketing it’ll die Contrary to present-day tech CEOs, the founding fathers of the field didn’t hide the enticing cover story that is the name “artificial intelligence.” Before I read Klein’s essay on Monday, I came across this LinkedIn post where Chris Wiggins, the chief data scientist at the New York Times, recalls that John McCarthy, the father of artificial intelligence, came up with the catchy term to get funding: “I invented the term artificial intelligence…when we were trying to get money.” It’s ironic that McCarthy said this to James Lighthill during the 1973 “Lighthill Debate” after the latter had submitted a critical report that led to the beginning of the first AI winter. McCarthy’s revelation didn’t cause the winter, though. The reason was the unachievable goals that the term “AI” implicitly (and explicitly) promised. The Lighthill report, as it’s known today, concluded that progress didn’t match expectations, causing the British government to withdraw funding from universities. The name, however, prevailed. “What are the arguments for not calling [the field] computer science … and calling it artificial intelligence,” Lighthill speculated in front of the audience. “It’s because one wants to make some sort of analogy. One wants to bring in what one can gain by studying how the brain of living creatures operate — this is the only possible reason for calling it artificial intelligence.” McCarthy then interrupted him and publicly admitted the motives that led him to coin “AI,” clarifying two critical truths; one that was immediately obvious and another that we’ve been feeling with increasing acuity for almost 7 decades: First, AI is a marketing term, not a scientific one. Second, although McCarthy’s reason wasn’t to “make some sort of analogy” with the human brain, as Lighthill presumed, it did cause this unintended effect. An anthropomorphism worth hallucinating over. Now, leading tech CEOs are making sure we don’t forget those two words. From marketing they were born, in marketing they will die. (Just to be clear, AI may be a marketing term but that doesn’t take away any of the characteristics that make it special: Early AI systems were greatly bounded by the limitations of the first computers but modern AI systems are wonders of science and engineering — in terms of design, creation, and human ingenuity and in terms of ability and performance. There’s no denying that even if we are to criticize who’s profiting from them, the flaws of the underlying sociopolitical system in which they’re deployed, or the false promises that accompany misguided concepts.) What if AI had a different name? Lighthill mentions “computer science” as an alternative to “AI,” but that’s taken. AI, however, is a bad name — what if it had a different name? When McCarthy was looking to launch the field in the early-1950s, Claude Shannon, the father of information theory and one of the founding fathers of AI, advised him not to use the name artificial intelligence because, in McCarthy’s words, it was “too flashy a term” that “might attract unfavorable notice.” They settled for “automata studies,” but it didn’t work out as expected. Eventually, McCarthy decided in favor of AI in 1956 during the “Dartmouth Summer Research Project.” He wanted to achieve the long-term goal of creating human-level general intelligence and that was, for him, the best possible name. Herbert A. Simon, also an AI pioneer, proposed a different one. Not slightly different, if you ask me, but possibly the opposite in terms of honesty and attractiveness: “complex information processing.” Much more accurate but definitely not very “flashy.” What would the world today look like if Simon got away with it and AI was instead called complex information processing (CIP)? This is pure speculation: My guess is that not everything would be better, as some of you may expect me to say. I can make the case for three differences for the better and three for the worse. For the better: First, CIP has no immediate connection to human intelligence. We wouldn’t suffer from anthropomorphizing terms like “hallucination.” Second, CIP sets an example of accuracy and honesty. If the field’s name isn’t an exaggeration, others won’t try to sell anything — CIP is a scientific term, not a marketing one. Finally, a consequence of the other two; without anthropomorphism or marketing AI hype wouldn’t exist. No one would see any potential gain from overhyping the field’s promises and expectations would match reality. For the worse: First, CIP isn’t attractive. Without a means to ignite the imagination of potential investors, they wouldn’t open their pockets (while we try to change the world for the better, marketing has a valid purpose). Second, CIP doesn’t point to long-term goals but to present (at the time) capabilities; it doesn’t express ambition. Finally, again a consequence of those two; without money, vision, or ambition, there’s hardly a path forward. I wouldn’t be writing this article. You wouldn’t be reading it. Leaving any political conclusions to Klein, I conclude here that albeit “AI” is indeed the original and greatest lie in the field (and we should be aware of that), not everything that stemmed from it was for the worse. It was a compromise that those who had to choose gladly accepted. Would we be better off had history been different? No one knows. No one will ever know. All we can do is make the most of what we’ve been given.
_ _ _ _ _ _ _ _ _ _ ► PEIRCE-L subscribers: Click on "Reply List" or "Reply All" to REPLY ON PEIRCE-L to this message. PEIRCE-L posts should go to [email protected] . ► To UNSUBSCRIBE, send a message NOT to PEIRCE-L but to [email protected] with UNSUBSCRIBE PEIRCE-L in the SUBJECT LINE of the message and nothing in the body. More at https://list.iupui.edu/sympa/help/user-signoff.html . ► PEIRCE-L is owned by THE PEIRCE GROUP; moderated by Gary Richmond; and co-managed by him and Ben Udell.
