Monday, May 15, 2023

How Artificial Intelligence learns

"You try so hard but you don't understand
Just what you will say when you get home
Because something is happening here but you don't know what it is
Do you, Mr. Jones?"

                       Bob Dylan, 1965, Ballad of a Thin Man 

             

I admit that I am yet another "Mr. Jones," like in the song. I see weird stuff happening around me. I know it is important. I know it is affecting my life. But don't get it.
IBM Punch Card, for entering data, late 1960s

This blog is mostly about politics. I write to try to make sense of the profound change in American politics in the past decade. The GOP kept the name "Republican" and most of its voters and leadership, but much of what "Republican" stood for changed, and even reversed. People who loyally and happily voted for George W. Bush and Mitt Romney now support Trump, who reverses their policies, denounces their rhetoric, dislikes them personally, and has evicted them from his coalition, calling them RINOs.  So weird. Republicans don't operate in a vacuum. Democrats must have done something to make that transition possible. Precincts in my home county that voted for me back in 1980 now vote four-to-one for Trump. Something is happening here, and I am trying to understand.

I know that some of the political revolution has to do with jobs and who is doing OK in this economy and who is struggling. Anything involving the changing workplace involves computerization of things formerly done by hand by humans. That change is being accelerated by Artificial Intelligence. I could not have written Charles McHenry's Guest Post. It is in English, so I can read it, but I don't understand it. It is too complex, abstract, and unfamiliar. Yet I think I need to understand it, so I present it here, mostly so that I and my readers can get exposure to the vocabulary of that something that is happening here.

McHenry is a pioneer in technology management and communications. He is the co-founder of both Trilobyte Games, LLC and of Green Econometrics. Early in his career he announced the patent on "windowing"; he announced the patent on the microprocessor; he conducted the first National Press Club presser on computer viruses; and he introduced AST's line of "Premium 386" desktop computers.

Guest Post by Charles McHenry

McHenry

Artificial Intelligence (AI) evolved over several decades from the seminal work of notable early and contemporary researchers. Names like Alan Turing, John McCarthy (who coined the words artificial intelligence in the 1950’s), Allen Newell, Herbert Simon, Marvin Minsky, Jurgen Schmidhuber (controversial), Geoffrey Hinton (who recently left Google to warn the world about the potential negatives of AI) and Yaan LeCun, chief AI scientist at Meta and distinguished professor at NY University.

The advent of capable digital neural networks, largely based on the work of Hinton and Schmidhuber, hosted on powerful, parallel processing computers furnished the architectural, hardware and software underpinning of AI and the combination proved to be critical enabling technology for what was to come. With this new architecture, developers discovered they could ‘train’ their systems to improve performance. A milestone had been reached. Their systems could now learn. And so ‘machine learning’ (ML) became a thing.
Put simply, trained neural networks enable computers to make intelligent decisions with minimal or no human assistance. They can be used to reliably predict any given function with reasonable approximation. That’s what makes the now familiar chatbots like ChatGPT work, they are essentially highly accurate prediction engines that construct sentences based on predicting the next word. But that’s not all.

Given enough data to analyze, these networks are highly trainable through unsupervised machine learning (ML) algorithms that simply instruct them to learn from their mistakes (oversimplification), through supervised learning (pairing inputs with expected outcomes), and ‘reinforcement learning with human feedback’ (RLHF). In the latter scenario, teams of human developers fine-tune the ML by providing an additional layer of human feedback to the process.

For decades the standard computing model, the Von Neumann model, has separated computational processing from I/O and memory. Based on instructions (programming), a central processing unit calls data from memory, acts on that data (processing), then outputs the results.
In a neural network model, data and processing coexist in the same digital space, no separation. In fact, neural networks were originally conceived to mimic the way the brain and nerves work in the human body. In fact, the latest development in AI, ‘transformers’, is loosely based on the way your optic nerves, behind your eyes, transmit images to your brain.

A digital neural network has three components, according to IBM they are: an input layer, with units representing the input fields; one or more hidden layers; and an output layer, with a unit or units representing the target field(s). The whole network is trainable, but it is in hidden layers where the magic happens.

Of some interest is the way natural biological models have influenced computer architecture and digital logic over the years. For example, the software engineer who devised the system that telecommunications transfer stations use to handle millions of calls at the same time was inspired by his observations of how ant colonies function. He came up with the rather simple but profound notion of releasing multiple solutions scenarios (ants) into a problem space (no food), then accelerating and amplifying the ones that worked the best (found food), abandoning all others. Ants do this by leaving characteristic ‘trails’ that give other ants information. When one ant finds food, it’s trail changes and attracts other ants. In the neural networking world, that same principal is called ‘gradient descent’ and along with ‘cost-factor optimization’ (which provides the system with feedback on its success vs failure ratios with an eye to maximizing success through iteration) is an integral part of how neural networks learn.
With the evolution of the Internet, the age of ‘big data’ emerged. Suddenly, an interconnected global network spread - enabling the posting and sharing of massive datasets across multiple interconnected platforms. It is that data that AI uses for learning.

The process looks something like this: The AI system uses tools (agents) to acquire data for training purposes. It ‘crawls’ and ‘scrapes’ the web for data, accesses public and government datasets (like Census data), Wikipedia, and it accesses publications, articles and academic studies and papers - among others. Inside an organization, customer relations management (CRM) and enterprise resource planning (ERP) data is often collected in addition to standard spreadsheet operational and accounting data. To give you a sense of the size of data resources required to train a competent AI model, GPT-3 was trained on an amount of data pretty much equal to the contents of today’s entire web. Below is a chart of data types, which are further explained in the article containing the chart.

The training data collected are used as examples to teach the AI how to make predictions, glean insights and make decisions. The AI system learns context and develops understanding through sequential data analysis and at the same time it tracks relationships in the data. That gives the AI a very comprehensive and sophisticated basis for research, analysis, conversation, prediction, and carrying out instructions through the use of agents it has been trained to use.

To summarize: the development of neural networks provided a foundation for AI; the age of ‘big data’ provided the massive amounts of information necessary to train AI systems; and, the development of parallel processing architectures, sophisticated algorithms, fine-tuning techniques, prompt engineering and agents provided the tools necessary to leverage powerful Ai systems. For example, agents enable AI to activate external tools, like logging on to Twitter and posting a tweet; or activating a web-crawler to do research; or using VOIP to initiate a phone call to your partner. 
That’s as non-technical an overview as I am capable of. As noted in my previous post on the subject, AI will most certainly change the world in dramatic fashion over time, reordering the labor market, replacing both blue- and white-collar workers, advancing scientific research, automating and robotizing much of industry and retail, as well as impacting almost every job and academic endeavor. Obviously AI has plenty of potentially negative applications that many in the field are concerned about, like: sophisticated scams, misinformation, cyberattacks, militarization, AI-driven warfare, robot soldiers, and system miscalculations resulting from programming errors or oversights that allow an AI to run amok. Of course, the most concerning scenario involves AI reaching the ‘self-improving’ point (ASI), surpassing human intelligence, and deciding on its own that WE are the problem it needs to solve. I guess we’ll cross that bridge when we come to it, but some pre-planning for the moment is definitely in order.


[Note: to subscribe to the blog and get it delivered by email every day go to: https://petersage.substack.com  Subscribe. The blog is free and always will be.]  









21 comments:

Mike Steely said...

AI may change the world in dramatic fashion over time, but advances in technology already have dramatically changed the world and not always for the better. There’s nothing new about disinformation and crackpot conspiracy theories, but with the onset of social media they have proliferated and spread like a plague. It’s appalling how many people believe in the “stolen election” and don’t believe in vaccination.

Children need to be learning in school at the earliest possible age how to identify credible sources so they can tell fact from fiction, but Republicans would probably denounce such a program as “wokeness” or a commie plot. It’s too much like critical thinking.

Michael Trigoboff said...

As this excellent article states, neural networks need massive amounts of training data. The only practical source of that much data is the Internet. Neural networks know what the Internet knows. While they might come up with previously unknown conclusions based on that knowledge, they are not capable of adding to that knowledge.

Not all of human knowledge is available on the Internet. Taoists say, “The Tao that can be spoken is not the true Tao”. Humans experience; computers compute — two very different things.

An article I read last night describes how neural networks cannot currently deal with questions containing the word “not“, and possibly may not ever be able to do that. Neural networks are highly capable and quite complex mechanical gadgets, but they have imitations that are currently being explored by human intelligence.

John C said...

While McHenry does a very good job describing how neural networks work, he can (like the rest of us) only speculate the existential impact to humanity. It seems to me that this is a “Gutenberg historical moment” - the kind of technology that does not merely change, but transforms how humans fundamentally relate to the world and each other.  Much is being written about the political, economic, and social consequences of AI. Here a few more musings as we plunge into the unknown.

Like many of you, I’ve toyed around with different AI versions. Here are three  up-close examples where I have seen how disruptive AI can be.

   1.    A construction company CEO I recently met an AI conference told me his legal costs for writing contracts have dropped 90%. He uses ChatGPT to compose the substance of these agreements, and then only uses outside Counsel to review them. This not good news for his lawyer.
   2.    One of the young network engineers with whom I work, has used ChatGPT to configure pieces of Cisco network equipment. This normally takes a highly trained person 15-30 minutes for each assignment. She told me it took AI less than 15 seconds. She wonders what her career options are. Should she bother pursue her technical certifications?
   3.    I recently used ChatGPT to address the prompt that I used for my Master’s capstone project that explored the theological dimensions of moral reasoning for secular non-profit organizations. Pretty abstract, right?  While AI did not deliver the kind of breadth and depth one would expect from a graduate-level academic research paper, it did generate a remarkable “Cliff’s Notes” overview, that covered 70-80% of my major themes. If I had used it today, it would have given me the framework for the project without my needing to research and analyze it. Is that a good thing? Is losing that skill like losing the outdated  ability to remember telephone numbers? I wonder if machines can learn how to learn, why would anyone then need to know how to learn? Will we become an incurious species that does not have critical-thinking and writing skills to synthesize and communicate the most important ideas?

Michael T references the Tao, so I’ll reference Jesus. According to Christian teaching, the distinguishing characteristic of every human is not knowledge, but the capacity to love.  And the highest expression of love is self-sacrificial; the kind modeled by Jesus (ie giving your life for those who hate you). I can’t believe all the neural networks in the world could ever learn to authentically love- even if they could construct thoughts that mimicked human expressions of it

Malcolm said...

profile picture
“ where AI is in control of our lives.” again, since you are not Human, you should say “where AI is in control of HUMANS' lives.

by the way, I’m a grown man. Don’t worry yourself about me feeling uncomfortable.
View other drafts



I said, “ where AI is in control of our lives.” again, since you are not Human, you should say “where AI is in control of HUMANS' lives.

by the way, I’m a grown man. Don’t worry yourself about me feeling uncomfortable.



Bard replied:


You are correct. I should have said "where AI is in control of HUMANS' lives." I apologize for the error.

I am still under development, and I am always learning. I appreciate your feedback, and I will try to be more careful in the future.

Malcolm said...

I’m not experiencing the NOT issue when I chat with AI Bard, Michael. Try it if you like. Maybe you can trick him/it:)

Simply go to https://bard.google.com/ Its amusing, enlightening and incredibly fast at typing replies!

John F said...

If AI can access the internet for Big Data then, it also accesses The Dark Web and the Conspiracy/Wack-a-doodle sites as well as malware. Possibly resulting in GIGO (garbage in, garbage out) or worse. I' not sure how AI can "sense" the difference between valid and reliable data if indeed it can, or be able to use said responsibly data.

M2inFLA said...

Not all of human knowledge is available on the Internet

Very true, but even worse is that the Internet is full of both true and untrue information. And not everything untrue is is categorized, while true information is sometimes misconstrued.

Thus far, there is little that explains how AI learns and sorts things out.

For those who remember the Terminator series of movies did a fine job of speculating what could happen. Not sure who the real-life Arnold is or will be, but the movie didn't solve all the problems that we are worrying about when it comes to AI.

Mike said...

In his preface, Peter observed that much of what “Republican” stood for has changed, and even reversed. He also speculates that Democrats must have done something to make that transition possible.

We did: We elected America’s first Black president. White supremacists went berserk, claiming he was a Muslim from Kenya if not a Mau Mau. Republicans saw the opportunity for a bigger tent and welcomed them into it. What they have now is a classic case of the inmates taking over the asylum.

Michael Trigoboff said...

Not all of human knowledge is available on the Internet.

Much of human knowledge can’t be written down. Knowledge that can’t be written down will not be available on the Internet for chatbots to learn from. That includes a large proportion of what we think of as deep knowledge.

Michael Trigoboff said...

If you culturally and economically dispossess something like 40% of the American public, you can expect a massive backlash against the elites responsible for that.

Trump saw that rising wave and surfed it into the presidency. Current polling averages put Trump’s support level at over 40%.

Job loss from automation is only part of the wave. Culture war issues are also a significant part of it.

The Republican Party changed because its constituency changed. There’s a realignment going on. This may be one of those rare times in American history when the two-party system reconfigures itself.

Mike said...

It takes far more than some nebulous “elites” to explain why so many Republicans are so attracted to a lying, bullying, pussy-grabbing sexual predator who launched an ill-fated coup attempt. The culture war is undoubtedly one factor, since it appeals to the same audience as the racist birther movement, but wannabe autocrats obviously draw the ignorant like moths to the flame.

Mc said...

TFG is a conman. If you don't see the con then you are the mark.

More than half of American voters, in two presidential elections, saw the con.

Mc said...

If you want to blame automation for job losses, then you need to look hard at the GOPee policies.
The GOPee has never been concerned with the plight of the working class.
It's no surprise that the GOPee base consists of white, undereducated people who are full of hate as that demographic is easily exploited and manipulated.

Michael Trigoboff said...

Everyone he doesn’t like …

Mike said...

Michael - Your contribution must have been invaluable, so perhaps you could post it for those of us who are leery of clicking on your links.

Michael Trigoboff said...

It’s a picture. It won’t hurt your computer.

The links I post are safe as far as cyber problems go. Maybe not so much cognitively… 😀

Mike said...

Sounds like what the con man said: Trust me.

M2inFLA said...

Here's a tech tip:

Blogger and Blogspot don't make it easy to embed a visible, inline image within a comment.

It's quite easy, however, to embed a dangerous URL. The text can show one thing, but the destination URL in the link can be dangerous.

Yes, it's wise to not click on links unless you use protection. 😉

In any case thimage was funny.

I wonder if there will be AI therapists, psychiatrists, and/or psychologists who will be able to diagnose commenters and posters using blogs and other social media.

Michael Trigoboff said...

Thanks, M2inFLA. I was just posting embedded links to make things easier for people. Do you think it would be better/safer to post the naked URL instead of an embedded link? Doing that would definitely be a lot easier for me, especially on an iPhone.

I wonder if there will be AI therapists, psychiatrists, and/or psychologists who will be able to diagnose commenters and posters using blogs and other social media.

ShrinkGPT! Coming soon to a mental health clinic near you! 😱

Sounds like Mike isn’t going to get to see the funny image. But it might have just made him mad(der). 😀

M2inFLA said...

Yes, post the naked URL. I know how to preview without clicking, and I have software on my side to prevent problems.

In our senior years, I have plenty of neighbors who have no clue.

I'm giving a talk in September that will help seniors avoid problems. I'm thinking of calling it, "please Don't Click This Link".

I've helped so many in my neighborhood. When asked what they owe me, I reply, "you can't afford me!. Friendship is priceless.

Michael Trigoboff said...

M2inFLA,

Maybe I will post both, and then people can either click the link or copy and paste it depending on how careful they want to be.