AI is everywhere. We can't see it, but it’s on nearly every app we use on a regular basis and websites we scan daily, such as Facebook, Twitter, and Google.
The beginnings of AI
The term Artificial intelligence is quite new, only having been around for the last century. Interestingly enough, though, AI is even older than electronic computers. But what is the origin of AI? The father of computer science, Alan Turing - for which the prestigious Turing award is named - spoke about AI even before WWII. He published a famous paper nicknamed “Can Machines Think?” in 1950, from which we've gained two things:
- A view on AI: Thinking is as thinking does.” The way you act reflects the way you think. You are hungry for a banana, so you eat the banana on your desk. Turing isn’t going too deep - philosophically or theologically - and assumes that thinking is behavior.
- A test: Let’s put someone into another room and see if they can tell the difference between a human and a computer. We will call it the same as if we were speaking about a human.
Although Turing’s paper set the tone for the years to come, the official coining of the term ‘artificial intelligence’ occurred in the summer of 1956 during a conference at Dartmouth. John McCarthy called together a dozen people to try and narrow down the kinds of problems that could be solved using computer programs.
Around this time, computers were growing up from nothing. By 1956, something really existed that you could run programs on. Computer science was starting to become a field in its own right.
From Numbers to Symbols
In the beginning years, computers were understood to only be capable of completing computations with numbers. But what about other things that can be computed? One of the first successful experiments was done in propositional logic by Simon and Newell.
In the 20th century, Bertrand Russel and Alfred North Whitehead had been investigating the foundations of mathematics and wrote Principia Mathematica on propositional logic.
Propositional logic is a field that studies mathematical statements using elements such as p, q, r, s. These are not objects in the real world. Propositional logic dates back to Aristotle, who originally fashioned propositional logic to model reasoning.
The experiment was an excellent choice because, at the time, computer science didn’t know how to connect programming to anything concrete in the world.
The program used in the experiment proved all the theorems in chapter two of the Principia Mathematica. The program also produced a shorter, more precise proof for one of the lengthy proofs done by Russel and Whitehead in that chapter. This impressed Russel, who responded, “Now, I’m prepared to believe a machine can do anything.”
AI in the hard sciences
Computer science could solve deductive proofs. Deductive arguments are arguments that, if the assumptions are correct, the conclusion necessarily follows. Reasoning follows this form and it’s one way of gaining knowledge from the world.
But we also know the world through experience. We also can observe events happening in the world and come to conclusions based on those observations, in other words, induction. The next transition computer science made was into the world of the hard sciences – into the physical world.
One early experiment was done in collaboration with a biochemist to help detect life on Mars. AI was used to help analyze mass spectrometry (MS) data. The results were impressive, resulting in findings that rivaled the work of postdocs at that time. Soon after, the field of computational biology was born.
Search and Knowledge
At the beginning of AI research, people quickly realized that the essence of problem-solving is search.
Problem-solving, broken down, is really a series of decisions. Searching for the best solution is implicit in the symbols you’re using. You are looking for a good enough answer to get to a solution. These decisions map well onto the decisions that a computer is making at the very core: zeros and ones. The number of possible decisions you can make is extremely large and search heuristics help you narrow those decisions down.
AI quickly broke down into two groups, one of which was trying to make AI as smart as possible, beyond the kind of intelligence found in humans. The other group was trying to imitate human behavior and use realistic models to guide development – to behave like psychologists.
Later, people noticed that programs became better and better the more information that was fed into the systems. And a shift into a new way of thinking happened.
“Through knowledge, comes power”
- Edward Feigenbaum
Expert systems and machine learning
Through the shift to knowledge, expert systems were born. The system emulates the decision-making ability of an expert. Expert systems solve rules using statements in the form of propositional logic (if-then) statements.
Around that time, a field called machine learning also developed. Machine learning is used to recognize patterns in a large amount of data and learn from it. This branch of AI merged with statistics to become the field that it is today.
The difference between machine learning and semantic AI can be compared to between the two ways humans think as laid out by Dan Kahneman in Thinking, fast and slow.
The AI we observe in our everyday apps has shifted from expert systems towards machine and deep learning. From the era of knowledge, we have begun to move into the age of computation.
There are several conclusions that can be taken away from this introduction to AI. Here are a few of them:
More data, better solutions.
- Knowledge, or data, means power. The results of our AI algorithms always improve with more data and correctly labeled data.
- In order to start using AI technology, you need data to create your AI, whether you’re thinking fast or slow.
- If data determines the quality of your AI, then make sure your data is high quality. Trace the path in which your data is being collected and know where your data is coming from.
Look for problems that can be solved by identifying patterns.
- A problem that requires quick decision-making or recognition of patterns is perfect for machine learning.
- We will delve more into machine learning, natural language processing, and other related concepts in the next article.
We don’t know if AI is conscious.
- People do not agree on what consciousness really means, which makes this question difficult to understand.
- Human-level consciousness hasn’t been achieved, yet.
The computation is the next problem to tackle in AI.
- Computational speed limits us at the present moment.
- Keep your eye on how quickly your AI can deliver. Real-time AI is a difficult point to reach.
- Could developments in 5G be useful?
The definition of AI will continue to develop over time.
- Where did AI come from? Turing created the original soft definition of AI. He asked if machines could think and proposed a simple experiment.
- AI is with computer science as a technique that intends to imitate human intelligence and behavior, or better.
If you want to hear more about this topic, please tune in to this podcast for the conversation between Edward Feigenbaum and Robert Harrison that inspired this article.