Okay, that’s ok, 2023 has definitely been “the year of AI”. But there have already been “years of AI”. At least one.
As this academic journal reports, there was a “year of AI” 43 years ago, in 1980. So AI has been around for a very long time.
Several decades ago, I wrote a thesis on the ethics of AI. In 1986, I wrote an article for the now defunct Computer Design Magazine called “Artificial Intelligence as a Systems Component.” Then, in 1988, I introduced two AI-based products for the Mac.
Even then, AI was already more than 30 years old. Some of the earliest activity in AI dates back to Professor John McCarthy, a teacher and researcher at Stanford and MIT. In 1955 he founded SAIL, Stanford’s AI lab, and in 1958 he invented the adorable LISP (one of my favorite programming languages of all time).
In 2024, AI will therefore have existed for at least 69 years. And that’s without counting fiction. Isaac Asimov actually considered the ethics of AI 25 years earlier, in 1940.
And yet, I would have a hard time objecting to calling 2023 the year of AI. Yes, it was an exceptional year.
What has changed in 69 years?
AI has been used for a very long time. Whether in expert systems, diagnostic tools, video games, navigation systems or many other applications, AI has been used for decades.
But it has never been used as it was in 2023. 2023 is the year when generative AI truly established itself. While many years could claim the title of “Year of AI,” there is no doubt that 2023 is the “Year of Generative AI.”
The big difference, the one that has led to the huge explosion of truly useful AI this year, is how we are able to train AIs. Until now, AI training was mostly supervised. In other words, each AI was fed specific information, which makes up the AI’s body of knowledge. This supervised pre-learning restricted what each AI knew, and what it could do.
On the other hand, today we are in the era of large language models (LLM), where prelearning is not supervised. Rather than introducing a limited set of domain-specific information, AI providers like OpenAI fed AIs with pretty much everything they could: the entire internet and pretty much any other digital content they could get their hands on.
This process allows AI to produce material that is astonishingly varied and on a scale that was previously impossible to achieve.
And this has been made easier by the considerable progress made in processor and storage performance. In 1986, you could get a hard drive the size of two microwaves and the weight of a refrigerator for $10,000 (about $27,000 in today’s money). It contained 470 megabytes. Not gigabytes, not terabytes, but megabytes.
Today you can get one on Amazon for 279 euros. The combination of the cloud, bandwidth, much faster processors in the form of CPUs and GPUs, and much larger RAM pools makes the processing power of LLMs possible.
An example in the field of horticulture
To give you an example of this difference, let’s use one of the products I featured a few years ago. House Plant Clinic was an expert system that was trained by a horticulturist. My other product at the time was the expert systems development environment, Intelligent Developer, used to create House Plant Clinic.
The process was laborious. Through a very long series of interviews, another engineer and I obtained rules, facts and best practices from the plant expert, then encoded them into the knowledge base. On the advice of the plant expert, we also had illustrations produced for situations where users would need a visual.
Screenshot by David Gewirtz/ZDNET
The extent of knowledge at House Plant Clinic consisted of what we had encoded into the expert system, nothing more and nothing less. But it worked. If you had a question and it fell within the knowledge we had encoded, you could get an answer and be sure it was correct. After all, the knowledge provided had been verified by a plant expert.
Now let’s take the case of ChatGPT. I asked ChatGPT the following question:
I have a houseplant that is sick. Ask me questions step by step, requiring only one answer per question.
ChatGPT did a good job asking questions about soil moisture, leaf condition, etc. Although he didn’t offer a picture, when I asked him to show me a picture of the pests, with their name, that can be found on a houseplant, I got a very more advanced:
Screenshot by David Gewirtz/ZDNET
That said, no one – not even Google – has any idea what a “KRIDEFLIT” is. As we have seen time and time again, generative AI has a veracity problem.
So while ChatGPT can speak confidently on almost any topic, our much older, expert system-based project had a much better chance of being accurate. One of them was created and validated by a real subject matter expert, while today’s chatbot generates insights from a gigantic pool of unqualified data.
The generative AI that we used in 2023 and will use in 2024 can do much more. But all magic has a price.
Pandora’s Box of Generative AI
Generative AI is extraordinary. In the past year, I’ve used generative AI to help me build an Etsy store, to help me create album art for my EP, to help my wife’s e-commerce business by creating custom social marketing images, to create a WordPress plugin, to debug code, to do detailed sentiment analysis, and much more.
But generative AI is not without its problems. It presents a serious problem of precision. You can’t trust what AI produces. Because she has been trained on a very broad body of knowledge, she is incredible. But because it has been trained on such a vast body of knowledge, it is polluted by what we humans write and publish.
And then there are jobs. Today, ChatGPT is acting like a particularly talented intern with an attitude problem. He’s useful, but only when he wants to be. But as this technology evolves, it will be able to address bigger problems with more nuance, and then we’ll have bigger problems.
It’s one thing to rely on AI to help me multiply my time. But when big companies decide they would rather save money and use AI services, many people will lose their jobs.
This trend will start with entry-level positions, as ChatGPT is essentially an entry-level worker. But then, three other trends will follow:
- There will be fewer and fewer experienced workers because there won’t be enough newbies to enter the job market.
- AIs will become more sophisticated, and companies will feel comfortable replacing $100,000-a-year workers with $100-a-month AI subscriptions – even if the work produced by the AI is not as clean, sophisticated, nuanced or precise as work produced by paid professionals.
- Work quality and output will decline, as will accuracy, which will have a ripple effect on the rest of the economy and society.
The good, the bad and the ugly
We started 2023 excited about getting ChatGPT to write a Star Trek story. By the end of 2023, we had a much better idea of the good, the bad and the ugly.
On the positive side, we now have a useful, if unreliable, personal assistant that can save us time, help us solve problems, and enable us to get more work done.
On the negative side, we have an existential threat to the employment of all knowledge workers that speaks to our collective zeitgeist.
As for ugliness, there is work to be done in 2024:
- Find a way to increase accuracy without weakening effectiveness with too many guardrails.
- Present useful information and illustrations without plagiarizing those whose work is under threat.
- Prevent the misuse of AI to alter elections and other nefarious activities.
- Receive data and generate results long enough for them to have real meaning.
- Moving to other media, like video generation, which are as amazing as image generation tools.
- Helping students learn without giving them an unbeatable way to cheat on their homework.
AI has flourished in 2023 like never before in its more than half-century of existence. Technology has opened the way to powerful tools, but also terrifying consequences. What do you expect, hope and fear for 2024? Let us know in the comments below.
Source: “ZDNet.com”