Eric Holloway

Senior Fellow, Walter Bradley Center for Natural & Artificial Intelligence

Archives

Yes, ChatGPT Is Sentient — Because It’s Really Humans in the Loop

ChatGPT itself told me there could be humans crafting its input. My tests indicate that that’s likely true
OpenAI, recently released a new AI program called ChatGPT. It left the internet gobsmacked, though some were skeptical, and concerned about its abilities. Particularly about ChatGPT writing students’ homework for them! also appears to be ominously good at answering the types of open-ended analytical questions that frequently appear on school assignments. (Many educators have predicted that ChatGPT, and tools like it, will spell the end of homework and take-home exams.) Kevin Roose, “The Brilliance and Weirdness of ChatGPT” at New York Times (December 5, 2022) The really amazing thing is ChatGPT’s humanlike responses. They gives an observer an unnerving suspicion that the AI is actually sentient. Maybe it is actually sentient. Wait, what? You heard

CAPTCHA: How Fooling Machines Is Different From Fooling Humans

Automated censorship is intended to protect against a tidal wave of spam but it could certainly have other uses…
Readers of Mind Matters News have likely heard of the iconic Turing test. Computer pioneer Alan Turing famously invented a test to determine whether a program could pass as a human. The gist is, if a program can fool human testers into believing it is a human, then the program is intelligent. Not everyone is convinced. Thing is, it doesn’t take much to fool us humans! Take Eliza , a program of only a few hundred lines, written in the 60s, which fooled many people into believing it was a real human therapist. But what if we flip the Turing test on its head? Instead of a test where a program tries to pass as human, we use a test that a program cannot pass, but a human can. For example, consider the CAPTCHA test we encounter on many websites. The term “CAPTCHA”stands for

AI Art Is Not “AI-Generated Art.” It is Engineer-Generated Art

The computers aren’t taking over the art world. The engineers are. Just the way engineers have taken over the music world with modern electronic music
Creativity is a mysterious thing. Our world economy is powered by creativity, yet despite the best efforts of our best engineers, creativity has not been captured by a machine. Until recently. With the new school of AI things have changed. We now have GPT-3 that can digress at length about any topic you give it. Even more remarkable, we have the likes of Dall-E, Midjourney, and Stable Diffusion. These phenomenal AI algorithms have scaled the peak of human creativity. AI can now create art that has never been seen before: The new artistic AI has become so successful the image social networks have become flooded with their artwork. Some communities have even banned the AI art. But the AI art is fun, and imaginative! One enterprising individual even won a fine art

How We Know the Mind Is About Information, Not Matter or Energy

The computer program’s world is one of binary 0 or 1 decisions but the physical world is one of many different shades of more or less
It’s really hard to picture the “mind,” isn’t it? You might think of wavy ghosts, or a spectral light. But nothing very definite. The brain, on the other hand, is very easy to visualize. Images and videos are just a Google away. That’s why it’s easy to assume that our brains are the entities that do our thinking for us. The brain is not only easy to image, it is physical. We can (in theory) touch it. Poke it. The brain even runs off electricity, just like your computer. But what makes a computer run Windows? It isn’t just the transistors on silicon wafers. It isn’t just the electricity coursing through the circuits. Windows itself is a ghostly being, like our mind. It is the structure of electrical signals in your computer. But the electrical signals

How AI Neural Networks Show That the Mind Is Not the Brain

A series of simple diagrams shows that, while AI learns faster than the human brain, the human mind tackles problems that stump AI
Recently, I’ve been arguing (here and here, for example) that we can use artificial neural networks (ANNs) to prove that the mind is not the brain. To recap, here is the logic of my argument: Premise A: neural networks can learn better than the brainPremise B: the human mind can learn better than a neural networkConclusion: the human mind can learn better than the brain, therefore it is not the brain This means if we can conclusively show the human mind can learn better than a neural network, then the mind is not the brain. For Premise A, I’ve argued that the differentiable neural network is a superior learning model compared to the brain neuron’s “all or nothing principle”. The neural network has a “hot” or “cold” signal that it can learn from iteratively,

Can Computer Neural Networks Learn Better Than Human Neurons?

They can and do; when artificial intelligence programmers stopped trying to copy the human neuron, they made much better progress
Neural networks are all the rage in computing these days. Many engineers think that, with enough computer power and fancy tweaks, they will become as smart as people. Recent successes playing games and predicting protein folds pour gasoline on the AI fire. We could be on the edge of the mystical Singularity, when humans and computers will merge and we become immortal gods. Or not. Neurons firing Let’s wind the clock back to the beginning of neural networks. In computer science terms, they are actually a very old technology. The earliest version, called a perceptron, (a single-layer neural network) was invented in the 1960s, inspired by McCulloch and Pitt’s early model of brain neurons. But, the perceptron was ignored for decades because Marvin Minsky (1927–2016) proved that

Artificial neural networks can show that the mind isn’t the brain

Because artificial neural networks are a better version of the brain, whatever neural networks cannot do, the brain cannot do.
“Minsky brilliantly portrays the mind as a ‘society’ of tiny components that are themselves mindless” – Simon & Schuster, 1987 What is the human mind? AI pioneer Marvin Minsky (1927–2016) said in 1987 that essentially “Minds are what brains do.” That is, the mind is the result of electrical waves cycling through the brain, as neurons spike and synapses transmit signals. But is that true? Can we test this idea? We can indeed, using artificial neural networks. One of the most popular approaches to artificial intelligence is artificial neural networks. These networks, inspired by an early model of how neurons fire (the McCulloch–Pitts model), consist of nodes, where each node is similar to a neuron. A node receives signals and then sends them to its linked

Google’s Chatbot LaMDA Sounds Human Because — Read the Manual…

What would you expect LaMDA to sound like? Whales? ET? I propose a test: “Human until PROVEN otherwise”
Recently Google employee Blake Lemoine caused a media storm over the LaMDA chatbot he was working on, that he claims is sentient (it feels things like a human being). A heavily edited transcript has been released that shows him and a collaborator having a very coherent conversation with LaMDA. Many have been quick to dismiss his claims about the chatbot’s sentience, accusing the Googler of falling prey to the Eliza effect: anthropomorphizing a probability distribution over words (thus believing that he is talking to a human). The accusation is that Lemoine generated a large number of dialogs, then edited down the exchange to create a coherent narrative. Google placed Lemoine on leave, technically for breaking the non-disclosure agreement (NDA) that he signed when he went

The Salem Hypothesis: Why Engineers View the Universe as Designed

Not because we're terrorists or black-and-white thinkers, as claimed. A simple computer program shows the limits of creating information by chance
In the fun-filled world of internet debate between creationists and evolutionists, we encounter the Salem Hypothesis: Creationists tend to be engineers. Many explanations have been offered for this phenomenon (apparently named after Talk Origins contributor Bruce Salem): engineers are closet terroristscreationists are trying to protect their fragile beliefsa desire to exert authorityengineers like simple black and white answers There’s a reason internet forums are not known for flattering character analysis! Anyhow, the true reason for the Salem Hypothesis is summed up in this graph. Read on to find out why. Engineers are more likely to be creationists because they are familiar with what it takes to design complex things for specific tasks. Which is exactly what we

Dawkins’ Weasel Program vs the Information Life Acquires En Route

To demonstrate what is wrong with fully naturalist assumptions like those of Richard Dawkins’ Weasel program, I developed Weasel Libs, modeled on Mad Libs
In his famous Weasel program zoologist and philosopher Richard Dawkins shows that the simple combination of random mutation and natural selection (Darwinian evolution) can produce the English sentence, “Methinks it is like a weasel”, in a short time period. The point of his program is to demonstrate that evolution can generate the complex, pre-specified DNA sequences we find in biology before the heat death of the universe. His argument sounds persuasive because both English sentences and DNA sequences are made up of symbols. Both can be randomly modified anywhere, and by cumulative selection, they can plausibly adapt to the environment in reasonably short order. Writers in English can learn to pen best-selling novels through trial and error and audience feedback. Evolution can

Why GPT-3 Can’t Understand Anything

Without long-term memory, human conversation becomes impossible
There is a mathematical reason why machine learning systems like GPT-3 are incapable of understanding. The reason comes down to the fact that machine learning has no memory. It is just probabilistic associations. If there is only a 10% chance of going off topic, then after just seven exchanges there is a greater than 50% chance the machine learning model has gone off topic. The problem is that when prediction is just based on probabilities, the likelihood of making a misprediction increases exponentially. A long-term memory is needed in order to maintain long-term coherence. GPT-3 is essentially a sophisticated Markov process. What is important about the Markov process is that the next step in the process is only dependent on the immediate previous step, or a fixed number of previous

AI Companies Are Massively Faking the Loophole in the Turing Test

I propose the Turing Test be further strengthened by presuming a chatbot is human until proven otherwise
Computer pioneer Alan Turing was posed the question, how do we know if an AI has human like intelligence? He offered his famous Turing test: If human judges cannot differentiate the AI from a human, then it has human-like intelligence. His test has spawned a number of competitions in which participants try to fool judges into thinking that a chatbot is really a human. One of the best-known chatbots was Eugene Goostman, which fooled the judges into thinking it was a 13-year-old boy — mostly by indirection and other distraction techniques to avoid the sort of in-depth questioning that shows that a chatbot lacks understanding. However, there is a loophole in this test. Can you spot the loophole? What better way to have a chatbot appear to be a human than if the chatbot is

Does Information Weigh Something After All? What If It Does?

At the rate we create information today, one physicist computes that in 350 years, the energy will outweigh the atoms of Earth
In the 1960s, IBM researcher Rolf Landauer (1927–1999) observed that if the logical information in a computational system decreased, then the physical entropy in the system must increase (Landauer’s Principle). This conclusion follows from the principle that the entropy in a closed system can never decrease. A decrease in the logical information corresponds to a decrease in entropy. And factoring in the principle that the entropy cannot actually decrease, the physical system itself must increase in entropy when the information decreases. This increase in entropy will result in the emission of heat, and a reduction of energy in the system. Now Melvin Vopson, a physicist at the University of Portsmouth, has taken Landauer’s principle to the next logical step. He applies

Soylent AI is…people!

OpenAI advertises itself as AI-powered, but at the end of the day, the system is human-powered
In the sci-fi movie, “Soylent Green,” the big reveal is that a food called soylent green is actually made from human beings, the catchphrase being “soylent green is people.” Likewise, as I discovered from a recent exchange with OpenAI’s GPT-3, “soylent AI is people.” GPT-3 is the product of AI company OpenAI. The company made headlines in 2019 with the claim that their AI model was too dangerous to publicly release. OpenAI is not a mere research company. While their publicly stated goal is fairly modest – “Aligning AI systems with human intent” – their CEO Sam Altman has bigger plans. He left his very successful role as president of Y Combinator, one of Silicon Valley’s most successful venture capital companies, to

Dawkins’ Dubious Double Weasel and the Combinatorial Cataclysm

Dawkins has successfully reduced a combinatorial explosion to a manageable problem...or has he?
In Richard Dawkins’ book, The Blind Watchmaker, he proposed a famous (and infamous) computer program to demonstrate the power of cumulative selection, known as the “Weasel program.” The program demonstrates that by varying a single letter at a time, it is possible to rapidly evolve a coherent English sentence from a string of gibberish. Evolutionary biologist Richard Dawkins, from Wikimedia Commons The way the program works is as follows: First, a sequence of characters is randomly assembled by drawing from the 26 English letters and the space. Then, one character is randomly reassigned. The resulting sequence is compared to the phrase from Hamlet, a quote uttered by Polonius: “methinks it is like a weasel.” For every character that matches, a point

Can Computers –- and People — Learn To Think From the Bottom Up?

That’s the big promise made in a recent article at Aeon
Tufts University biologist Michael Levin and Columbia University neuroscientist Rafael Yuste have an ambitious project in hand: To explain how evolution “‘hacked’ its way to intelligence from the bottom up,” that is, from nothing. They base their thesis on computer science: This is intelligence in action: the ability to reach a particular goal or solve a problem by undertaking new steps in the face of changing circumstances. It’s evident not just in intelligent people and mammals and birds and cephalopods, but also cells and tissues, individual neurons and networks of neurons, viruses, ribosomes and RNA fragments, down to motor proteins and molecular networks. Across all these scales, living things solve problems and achieve goals by flexibly navigating different spaces –

Is AlphaZero Actually Superior to the Human Mind?

Comparing AI and the human mind is completely apples and oranges
The Google-backed AI company DeepMind made headlines in March 2016 when its AlphaGo game AI engine was able to defeat Lee Sedol, one of the top Go players in the world. DeepMind followed up this great achievement with the AlphaZero engine in 2017, which made the remarkable achievement of soundly beating AlphaGo in Go as well as one of the world’s best chess engines in chess. Lee Sedol, picture from Wikimedia Commons The interesting difference between AlphaGo and AlphaZero is that AlphaGo uses databases of top human games for learning, while AlphaZero only learns by playing against itself. Using the same AI engine to dominate two different games, while also discarding reliance on human games suggests that DeepMind has found an algorithm that is intrinsically superior to the

“Slightly” Conscious Computers Could Doom Atheism

That might sound surprising but let’s follow the logic of the “consciousness” claim through to its inevitable conclusion
Recently, Ilya Sutskever, co-founder of OpenAI, proposed that artificial intelligence (AI) may currently be “slightly” conscious. His claim was probably in reference to the GPT-3 AI that can generate text from a prompt. I’ve played with a couple of the linguistic neural networks a bit, and you can try them out here. Some of the output is quirky, which could be mistaken for personality and make the algorithm appear conscious. The algorithm also generates emotional statements, that can generate empathy in a human user of the system. Just as kids make believe their dolls are alive when they develop an emotional bond with their toy, the algorithm text generates empathy in the human user. It can make us feel a bond with — and anthropomorphize — the algorithm. Here

Chalmers and Penrose Clash Over “Conscious Computers”

Philosopher Chalmers thinks computers could be conscious but physicist Penrose says no
Two authors I’ve been reading recently are Roger Penrose and David Chalmers. Penrose is a physics Nobel laureate who has stoked controversy by claiming in The Emperor’s New Mind: Concerning Computers, Minds and The Laws of Physics (1989) that the mind can do things beyond the ability of computers. Chalmers is a philosopher of science who claims in The Conscious Mind: In Search of a Fundamental Theory (1997) that consciousness cannot be reduced to physical processes. Both thinkers are well respected in their fields, even though they articulate positions that imply that the mind’s operation is beyond current science. At the same time, they believe that there is a way to see the mind as part of nature (that is, naturalistically), albeit from a yet-to-be-discovered point

Are the Brain Cells in a Dish That Learned Pong Conscious?

Human-derived organoids learned faster than AI and always outperformed mouse-derived organoids in terms of volley length, raising troubling questions
Recently, science media were abuzz with a remarkable story about minibrains (mouse and human brain cells in a dish) learning to play the video game Pong: Scientists have successfully taught a collection of human brain cells in a petri dish how to play the video game “Pong” — kind of.Researchers at the biotechnology startup Cortical Labs have created “mini-brains“ consisting of 800,000 to one million living human brain cells in a petri dish, New Scientist reports. The cells are placed on top of a microelectrode array that analyzes the neural activity.“We think it’s fair to call them cyborg brains,” Brett Kagan, chief scientific officer at Cortical Labs and research lead of the project, told New Scientist. Tony Tran, “Researchers teach human brain cells in a dish