AI (Artificial Intelligence) and RI (Real Intelligence)



 

By Dennis Schuchman


 

Everyone is saying that AI is about to change the world, and they’re right. After decades of missteps and false promises, AI is coming of age and it will change the world in ways that I can’t even imagine.

 

I’ve been in the computer industry for over 40 years now, but I’m not a computer scientist. In fact, I’ve never even taken a computer science course. I’ve learned a number of programming languages, how various operating systems work and how to configure hardware and software systems and networks to create computing environments that let users get their work done. So, I’m not a scientist, but maybe more of an engineer. Or maybe just a mechanic. When it comes to AI, then, I’m just as much an outsider as anyone. Well, almost.

 

I think that the way the media present AI gives people the wrong idea of what it’s about. People seem to think that AI combines the best of human brilliance with cold, hard computer logic.

 

Wrong.

 

Let me tell you a story.

 

There’s a weekly podcast I like called Skeptics’ Guide to the Universe. The host is Steve Novella, neurology professor at Yale. He and his panel discuss science news and debunk pseudoscience. One of the regular features of the ‘cast is called Science or Fiction. Steve presents his panel with three science news items. The catch is that one of them is fake and the panel has to figure out which one it is. Sometimes there’s a theme; the news items are all on the same general topic. A recent Science or Fiction had a theme, but Steve was going to keep it secret until the end.

 

I’ll spoil the ending -- the theme was ChatGPT. Steve asked ChatGPT if it was familiar with the podcast and with Science or Fiction. The program apparently gave answers that, if you’d heard them from a person, would lead you to think that the person understood Skeptics’ Guide in general and Science or Fiction in particular. So, Steve took the next logical step of having ChatGPT put together that week’s Science or Fiction.

 

When Steve revealed the fake story, I thought, “Huh. I thought that was true.” Turns out I was right. Turns out all three stories were real and none were fake.


So how could a seemingly intelligent program get something so simple so glaringly, obviously wrong? And that brings up some other questions that are just as important. How does AI ever get it right? And how do we actually know it’s right? The answer, basically, is: we don’t.

 

I’m an old school programmer. Any program I write is based on an algorithm, a step-by-step procedure for solving a problem. Writing a program means translating the algorithm into whatever programming language I decide to use.

 

Let me give you an example. When I was a kid, my father showed me a little geometric figure. He asked if I could draw it without taking the pencil off the paper and without drawing over any lines. Not having a lot of patience, I didn’t find the answer, so he showed it to me. Many years later, after I’d been working with computers for quite some time, something (I don’t remember what) made me think about Dad’s little puzzle. It was obvious that there was more than one way to do it and it made me curious about just how many solutions there were.

 

So I wrote a program to find them all. My programming technique was what we call “brute force”. It basically tried every possible way to do it. Luckily there was a small enough number of combinations that the program could run pretty quickly. And, since I wanted to know if the program was working right, I had it print out every step of every attempt. I could check to see if the solutions it found were right by picking up a pencil and trying them. If any of the solutions were wrong or if the program seemed to be missing correct solutions, I could go back and look at all of the attempts. If I did that, I would find one of three things:

 

1. Nothing was wrong and I was just mistaken.

2. Something in my algorithm was wrong and led to a wrong

    answer.

3. The algorithm was right, but my coding was wrong.

 

Early AI tried to come up with algorithms that that captured the way humans think. But that led to limited success, so modern AI doesn’t work that way. It works via neural nets, which are an entirely different way of simulating how human brains learn things and solve problems.

 

Suppose we wanted to create a neural net that could look at a picture and decide whether or not it was a picture of a horse. The neural net would have an input layer which would consist of the pixels of a digital image being examined. The pixels are called input nodes. Then there would be an output layer consisting of two output nodes, which we can call Horse and Not Horse. All of the input nodes are connected to all of the output nodes. It’s called a neural net because the nodes are analogous to neurons and the connections to synapses. After looking at a picture, we would want the neural net to light up either the Horse or the Not Horse node. To start with, the network has to be trained – it’s shown a set of images that are known to be either Horse or Not Horse. What training the net does is to change the strengths or weights of the connections between each input node and each output node. At the end, an output node lights up if the sum of all of its weighted inputs exceeds a threshold value. Needless to say, it’s a lot more complicated in practice, but that’s the essence of it.

 

I don’t know if a neural net as simple as the one in my example could really solve the Horse/Not Horse problem and most neural nets today are made more complicated and more capable by adding hidden layers of nodes between the input layer and the output layer. Each node is connected to all of the nodes of the next layer. You may have heard the term deep learning. All that refers to is neural nets with multiple hidden layers. 

 

Unlike my little geometry program, which can print out the result of each step in the algorithm so we can understand exactly what it did, the only thing we can learn from a neural net after training is what the weights of the connections ended up being. And that doesn’t really tell us why those particular weights gave us the answer. In fact, you can start with two identical neural nets and train them with different sets of images and you’d most likely end up with two nets with similar accuracy, but with different weights for the connections. And they’d probably get different ones wrong.

 

Large language models (LLM’s) like ChatGPT are built on neural nets and add another layer of complexity -- generative AI. According to Wikipedia:

 

Generative AI models learn the patterns and structure of their input training data and then generate new data that has similar characteristics.

 

What’s missing here? LLM’s don’t necessarily evaluate the input data to decide if it’s true or false or relevant or irrelevant or biased or honest. And they don’t check the answers for validity. That’s why ChatGPT didn’t know that its Science or Fiction was wrong.

 

This means that AI has some of the same characteristics as RI (real intelligence), namely fallibility and inscrutability. There’s ongoing research trying to eliminate, or at least reduce, the fallibility and inscrutability, but it’s not there yet.


Don’t get me wrong – I’m not trying to diminish what AI has accomplished and what it will accomplish in the future. If you want a great example, go look up AlphaFold, which has solved a major problem that many, many scientists have been beating their heads against for decades with limited success. That solution will likely lead to a whole new generation of medicines.

 

But we just need to be aware of AI’s limitations. Would I trust AI to drive a car better and more safely than a human being within the next several years? Absolutely. Would I trust one to decide whether or not to launch a preemptive nuclear strike? Not any time soon.

Comments

Post a Comment

Most Popular Posts

Adam Smith's Pin Factory

Image
Adam Smith - Our Founding Father ADAM SMITH VISITS A PIN FACTORY   Adam Smith’s description of a pin factory is on the first page of  The Wealth of Nations .  (Chapter 1 – “Of the Division of Labour”)  Drawings of pin factories of this period show workers using hand tools. Smith says the process can be broken down into 18 distinct steps, including packaging the pins. Smith mentions that pin factory workers were poorly paid, despite their high productivity.    Adam Smith says he visited a pin factory employing 10 men who produced 48,000 pins per day.  If each of the ten workers had done all the steps themselves, Smith says each worker could produce only 10 or 20 pins per day.  So the pin factory replaces 2,400 to 4,800 pin makers. The increase in labor productivity (output per person per day) is as high as 50 times that of individual pin makers.     This reduction in unit cost or average cost (AC) and the huge i...
Read more

Bilateral Oligopoly

Image
The best movie about the Gunfight at the O.K. Corral is titled My Darling Clementine. A great western - great cast, great photography.  DEFINITION OF A BILATERAL OLIGOPOLY Models of market structure assume that the demand side is represented by a large number of buyers.  The structure of the market, and the assumed market outcomes, depends on the number of suppliers and how they compete.  Suppliers either post one price or exploit their knowledge of buyer categories by using price discrimination.  Despite the comments about the key role of consumers in determining economic performance, consumers are fairly passive when the discussion focuses on imperfectly-competitive industries.  But many markets are not structured this way; the buyers are not passive consumers but large corporations that do not passively accept suppliers’ prices.  Net prices are actively negotiated.  These markets are often bilateral oligopolies. In a bilateral oligo...
Read more

The Stock Market Crash of 1929 and the Beginning of the Great Depression

Image
Famous Headline Introduction The usual reasons given for the Great Depression – the stock market crash of 1929 and the later collapse of the banking system – do not tell the whole story. Available economic data indicate (there were no national income accounts in 1929) that a recession had already begun before the stock market crash. The crash of October and November of 1929 was a catalyst that made the recession worse but the partial stock market recovery in early 1930 did not end the recession. Industrial production continued to fall quickly and unemployment rose rapidly in 1930. Continuing farm and businesses failures over the next two years wiped out thousands of small rural banks and threatened total financial collapse in early1933.  For a fuller explanation, we have to go back to the 1920s to see additional reasons for the origins and the rapid decline in output at the beginning of the Great Depression. We have to look at the int...
Read more

List of Posts By Topic

FIND POSTS BY CATEGORY CATEGORIES Basic Concepts and Theory Market Behavior and Structure Market Dynamics and Information:  How Markets Work The Industrial Revolution in England American Economic History American History Demographics and Economics Management World War I:  The Beginning of the 20th Century The Roman Republic and America - Differences and Some Possible Parallels Economic and Fiscal Policy Financial Markets and Investment Strategies Foreign Exchange Markets The United States American Foreign Policy and International  Relations Geopolitics and the Global Economy Geopolitics of Oil and Natural Gas Visionaries BASIC CONCEPTS AND THEORY Introduction to Economic Theory Economic Development and Economic Growth Production Functions and Supply Chains Demand  Analysis The following post was contributed by Dennis Schuchman. Artificial Intelligence (AI) and Real Intelligence (RI) You, Your Brain and Credit Card s Critique of Basic Economic Theory MAR...
Read more

Explaining Derivatives - An Analogy

Image
You go around to farmers with cows. You buy all the cows and pay the farmers a small fee to milk the cows and sell the milk. You pay for the cows with ass(et)-backed securities called MBSs (Milked Bovine Securities) that you tell investors are udderly safe. But some of the cows don't give enough milk (cow flow problem) or give no milk at all. You take some of the asset-backed securities, say they're backed by the subprime cows, and use them as collateral to sell another set of securities called CMOs (Cow Milk Obligations). Then you buy CDSs (Cow Dried-up Swaps) from AIG (Angus Insurance Company) to insure the CMOs when the cows stopped giving milk. If you work it right, you collect more on the CDSs than you pay out to retire the CMOs. The money you get from selling the dead cows go to pay the CLOs (cow leather obligations). You could also sell CDOs (cow dung obligations) that depend on how much cow dung is produced. This is a typical Wall Street product - turning sh...
Read more

Government Finance 101. Fiscal Policy: Welcome to Alice in Wonderland

Image
  PRELIMINARY SUMMARY OF FISCAL 2024 BUDGET The deficit in fiscal 2024 was around $1.8 trillion, about $5.0 trillion in revenue and $6.8 trillion in expenses. Interest on the national debt this year is around $960 billion, over twice the interest on the fiscal 2021 budget. Interest expense this year passed outlays for the military. Leaving aside Social Security and Medicare, interest expense is about 20% of total budget outlays. Interest expense is about half the budget deficit.   The fiscal 2024 deficit was 6.4% of GDP.  SOME BASIC DEFINITIONS AND A LITTLE DETAIL Some basic definitions for people in government suffering from political amnesia:   Deficit.    The difference between the federal government’s spending and its revenue in one fiscal year.  The fiscal year starts on October 1.  So fiscal year 2024 started on October 1, 2023. You know right away this is going to be confusing. Debt .  Short for national debt or ...
Read more

John von Neumann Sees the Future

Image
John von Neumann JOHN VON NEUMANN CREATES THE FUTURE If you want to understand the world we live in, you have to know about the thinking of three men – Alan Turing, Claude Shannon and John von Neumann.   I’ve talked a little about Alan Turing in an earlier essay.   All three men knew each other, discussed their ideas with each other and promoted each other’s ideas. John von Neumann has been called the smartest individual of the 20 th Century.   Even an outline of his accomplishments, which you can read about in  Wikipedia, is hard to believe.   He invented game theory, which I discussed in two earlier posts.   He headed up a group of mathematicians who invented new statistical techniques at Los Alamos that made the atomic bomb possible.   In 1944, he wrote a memo that outlined the structure of the modern computer.   He then managed the design and building of a modern general computer and consulted on the building of virtually all ...
Read more

The Roman Republic Commits Suicide: A Cautionary Tale for America

Image
Roman Legion Why Study Rome? Americans have focused on comparisons between Rome and America twice in American history.  The first time was during the period from before the Revolution to after the ratification of the Constitution and Bill of Rights, when the founding fathers and others looked to the Roman Republic for ideas and inspiration on how to structure the new government. After the American Revolution, the founding fathers, having read the Roman authors detailing the collapse of the Roman republic, worried about internal threats to the survival of their new republic.   The second time was when the United States became a superpower, especially after WWII, and then again after the collapse of the Soviet Union in 1991. Commentators compared America to the Roman Empire at the height of its power. There was some discussion about how America became a permanent warfare state and how it affected domestic democracy and civil society. ...
Read more

Josiah Wedgwood, the Wedgwood Pottery Company, and the Beginning of the Industrial Revolution in England

Image
Wedgwood Anti-Slavery Cameo A better introduction to the beginning of the Industrial Revolution than Adam Smith is the history of Josiah Wedgwood and the Wedgwood Pottery Company. This is how one company actually ushered in the Industrial Revolution. The revolutionary generation that first adopted steam engines saw the following trends and changes: Manufacturing was being modernized by a small group of entrepreneurs. Much of the new raw material processing and manufacturing was concentrated in a small area in the middle of England, away from London. These modernizing entrepreneurs formed a new economic, intellectual and social network. Modernizing entrepreneurs like Wedgwood tended to be members of Dissenting sects (like Quakers) or Nonconformist churches (not members of the Church of England), Whigs (liberals) in politics, and believers in “progress.” They were optimistic about the future, influenced by the ideas of Hume, Rousseau, Locke and Adam Smith. They believed...
Read more

“Pax Americana”: The World That America Made

Image
There are two kinds of empires: the kind that rules through ruthless exploitation, and the kind that seeks to induce allies into defending themselves through economic incentives. (George Friedman) INTRODUCTION America's power and influence in the world rests on three broad, interrelated bases: 1) Overwhelming military power that can be projected anywhere in the world,   supplemented by security arrangements with allies and supporters. 2) A large, innovative, dynamic domestic economy as the basis for global economic and technological leadership. The economy provides the resources and technology to support the military and a global foreign policy. 3) Claims to moral and ideological leadership and influence, particularly the promotion of liberal democratic societies and supporting political institutions. This post will discuss the economic aspects and related security objectives of American foreign policy. It will focus on the structure of the global...
Read more