The Brave New World of Artificial Intelligence By William Lama, Ph.D

My friends who meet at Mayers bakery are quite interested in artificial intelligence. We are the “Bakery Boys” and each week we solve one of the world’s big problems. Unfortunately, by the next meeting we have forgotten the problem and the solution. No wonder we are interested in artificial intelligence, or AI.

But some people are very worried about AI. Will it spy on us, eliminate our jobs, or worse? Ten thousand people have signed a Future of Life Institute petition that demands a six-month moratorium on AI research.

Even tech gurus are worried. Elon Musk told Tucker Carlson:

"AI is more dangerous than, say, mismanaged aircraft design or production maintenance or bad car production. In the sense that it has the potential, however small one may regard that probability, but it is non-trivial, it has the potential of civilization destruction."

Elon Musk tells Tucker potential dangers of hyper-intelligent AI - YouTube

I am no Elon Musk, but I like Ed Clint’s sober assessment: “We are the shapers of whatever AI we wish to make. This means we can expect them to be more like our favorite domesticated species, the dog. We bred dogs to serve us and we will make AI to serve us, too.” (Edward Clint, “Irrational AI-nxiety,” Quillette 12/14/17) Then again, if you saw me following Charlie the Westie around, you might wonder who is serving whom.

Current AI

Two years ago, I wrote a piece about Artificial Intelligence (AI). In it I focused on some of the things that AI could do.

It’s a whole new world based on Artificial Intelligence (AI) By William Lama, Ph.D.

I also discussed the new computers needed for AI applications. These included “Neuromorphic” computers that mimic the physical structure and signal processing techniques of biological brains by arraying the processor and memory units in neural networks. One application is autonomous vehicles where the AI system must incorporate the expertise that humans develop as experienced drivers.

Neuromorphic Computing - Next Generation of AI (intel.com)

A lot has changed in the last two years. AI-bots now pass the famous “Turing test.” We cannot tell if we are talking to a “chatbot” or to another human. “Hello Alexa.” 

“ChatGPT” is an artificial intelligence chatbot released in November, 2022. I asked it to tell me about AI, saying I was from India. Here is the (lightly edited) response.

Namaste! I am ChatGPT, an Artificial Intelligence model.

Artificial Intelligence (AI) is the intelligence demonstrated by machines. It involves creating machines that can perform tasks that usually require human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. It means making machines smart enough to do things that typically only humans can do.

AI includes machine learning, which enables machines to learn from data and improve their performance over time, and natural language processing, which allows machines to understand and respond to human language. (Siri is Indian.)

So that was my explanation of AI. I hope it was helpful! Dhanyavaad!

ChatGPT can even write poetry. Here is a poetic interpretation of “quantum entanglement.”

 

In the world of atoms and particles small,

A strange phenomenon does enthrall,

A connection between two particles unexplained,

As if their fates are forever chained.

Quantum entanglement is the name,

A concept that puts our minds in a game,

Two particles, no matter how far,

Their properties entwined like a guitar.

…….

Ok, not exactly Shakespeare, but ChatGPT is still a kid.

……

These examples of ChatGPT functionality are trivial compared to its full capability. For example, here is a dialog regarding construction of a “Dyson Sphere” (to capture energy from a star) including Q/A about materials, sources of the materials in the solar system, and the estimated cost. It is amazing.

Building a Dyson sphere using ChatGPT - 3 Quarks Daily

Artificial Intelligence seems to be everywhere, from voice and facial recognition, to factory and home automation, computing and chatbots. In many other areas expect quantum leap improvements from AI, including scientific modelling, auto-driving vehicles, and personalized medicine. 

 Future AI

Despite the wild success of ChatGPT, the current artificial neural networks are limited by their basic building blocks - individual artificial neurons.

A biological brain neuron receives input signals from other neurons through dendrites. The nucleus produces an output signal based on the input signals. The output signal is carried away by the axon to the next neuron.

An artificial neuron - a “perceptron” - receives its data from other perceptrons. The nucleus of a perceptron (colored in blue) performs calculations based on the input values and the activation function produces the output value, which will be the input for the next perceptron.

To see the basic problem with these artificial neural networks, consider one that distinguishes circles and squares. You would need two neurons, one for circles and one for squares. If the objects could be two colors, say blue or red, you would need four neurons: blue circle, blue square, red circle, and red square. More features mean more neurons. According to Bruna Olshausen, “You have to propose that, well, you have a neuron for all combinations. So, you have in your brain a purple Volkswagen detector.” Imagine how many neurons you would need for the auto industry. 

Bruno Olshausen | EECS at UC Berkeley

Olshausen argues, instead, that every piece of information in the brain is represented by the activity of many neurons. The perception of a purple Volkswagen is not encoded by two neurons, but by thousands of neurons. This same set of neurons, firing differently, could represent a pink Cadillac, a yellow Volkswagen, a silver Tesla or millions of other combinations. In our AI systems we would want our perceptrons to work that way.

This is the starting point for a radically different approach to AI computation known as “hyperdimensional computing.” (“hyper” meaning many, many) In this approach, all the salient features of an object are represented by a single hyperdimensional vector, a string of ordered numbers. Leading to bigger and better AI in our Brave New World.

A New Approach to Computation Reimagines Artificial Intelligence | Quanta Magazine

As much as I like the pet-Westie image of AI, I cannot escape Musk’s dystopian view. When the latest chatbot version, GPT4, was asked about the dangers of AI here is part of the response:

It could also be used to create deceptive or harmful content that exploits human biases, emotions, and preferences. This could include fake news, propaganda, misinformation, deepfakes, scams, or hate speech that undermine trust, democracy, and social cohesion.

And what might AI do about an existential crisis?

“A lot of data, from the most respectable sources, says that the world is threatened by man-made climate change. The obvious solution to that problem must be to decimate or wholly eradicate homo sapiens.”

Goodbye, world.

Artificial Intelligence: The Aliens Have Landed and We Created Them - Bloomberg



William Lama has a PhD in theoretical physics from the University of Rochester. He was a college physics professor and a scientist at Xerox Research Labs. He spent his last decade at Xerox managing software and electronics R&D. After retiring he served as Palos Verdes Library trustee for eight years, three as president of the Board. He may be reached at wlama2605@gmail.com.


RELATED ARTICLES

〰️

RELATED ARTICLES 〰️