From the moment the first computer was commercialized, people started asking themselves what computers could do for us. It was notably evident how powerful those machines were at helping us find information, provided an accessible platform for arcade gaming and facilitated typewriting. Then one day, during Summer 1956, scientists gathered for a workshop held on the Dartmouth College campus. The main intention of the workshop was to advance the computers’ capability. During this event, some even predicted that computers would be able to reason humanlike within the span of one generation only. It turns out that this could not have been more wrong. Now, after two whole generations, we are still waiting for such a technology to surface.
The slow rate of development is not only due to the computing power often not being strong enough to support the most complex algorithms, but it is also because of the recurrent fading in popularity. Every time expectations are proven too hard to overcome, this field of research notices a substantial drop in fundings. These financial pitfalls have caused the progress to slow down and have deterred newcomers to get their hands in the game. In this decade, however, we can notice that we are back to a peak of interest. The arrival of deep learning and machine learning has reignited the flame that once existed of producing a machine that would legitimately be humanlike. Yet, I don’t believe this is at all reasonable to expect such things. My training in neuroscience has shown me on many occasions how complex the human mind and behaviours are.
So, before hoping for an Ex-Machina to be produced or a Jarvis (Iron Man) to be made, we must understand what the bases of artificial intelligence are. Let’s start by defining artificial intelligence (AI). Well, the first part is by far the easiest to explain. We could effortlessly replace artificial with the synonymous word ‘man-made’; the second part, however, is much more complex. For a better and more complete representation of intelligence, I suggest you read my post from last week (Bring Up Intelligence). In sum, I define intelligence as being multimodal with as much as ten possible intelligence. Anyhow, what most people mean when they talk about intelligence in AI is either one, or a combination, of the four approaches.
The first is acting humanly, where one AI reacts similarly given the same situation and environment. The second is thinking humanly, where the AI would produce the same sort of thinking we humans do, with all its flaws and biases. The third is acting rationally, where the AI behaves following rules and seeks to find solutions, even though the results could be proved useless given certain conditions. The fourth, and last, is thinking rationally, where the AI forms preset notions and guidelines on how to respond given different circumstances. Regardless of which combination you choose, the result won’t be anything like the AI pictured in most Sci-Fi movies. The news channels do not make it easier either. News anchors may tell us that the best chess players were all beaten by AI. Some others may report that the longest winning-streak player of Jeopardy! was defeated by AI. Yet nobody knows if they genuinely qualify as AI. Some scientists are indeed entirely dismissing them as AI on the basis that they do not think and thus cannot understand the games. They only react based on their learnt knowledge and the strategies they developed.
So, it becomes hard to say when AI will realistically be intelligent. Moreover, if you consider the multimodal view of intelligence, then it becomes almost unfeasible. Gardner’s theory of multiple intelligence suggests that there are officially up to nine types of intelligence. Trying to recreate them artificially has met many successes but a lot more failures. For example, reproducing logical-mathematical intelligence has been highly successful. But every attempt at recreating creative intelligence has shown very miserable results. If we bring them all up in consideration of their reproducibility potential, we have: logical-mathematical, bodily-kinesthetic, visual-spatial, interpersonal, linguistic, intrapersonal, creative, musical and spiritual. Let’s note that most, if not all, of those successes, failed to introduce more than one intelligence type. So, trying to bring them all up in one machine to create a proper AI has still a long way to go.
One way to avoid dealing with all these different intelligence types would be to omit them entirely and instead opt for a criteria-based description of true intelligence. However, coming up with all the various criteria can form a notion that can sometimes feel incomplete. So trying to encompass all the elements forming intelligence is a serious exercise. The most recent attempt at determining these criteria brought up elements like reasoning, understanding, determining or detecting lies, finding relationships between items, considering meanings, and separating facts from beliefs. And all these components make sense. Getting to analyze a situation and coming up with explanations requires both reasoning and understanding, but these skills alone don’t make you particularly smart. Being able to pick your knowledge carefully is also critical. But let’s be honest, some information may need some special treatment as it may also hold very different meanings giving varying context. As a final touch, being able to assemble all of our knowledge together is definitely an advantage.
However, even with this new definition, we can’t realistically produce something that will include all of these components. Right now, all attempts at building AI have been separated into two groups: strong AI and weak AI. Strong AIs are unspecialized and thus can theoretically do many things, but they produce very weak results. For better outcomes, weak AIs are actually the way to go, even though they are highly specialized and thus, really know how to do one thing only. One downside to consider for weak AI is that they will never be independent. They will always need an external actor to compensate for their lack.
AI development made lots of progress in the last few decades, regardless of the difficulty we have faced defining what does or doesn’t classify as AI. The quest began with a handful of expert systems that were basically merely following set operators. These expert systems are computer programs following rules based on different algorithms. For instance, the grammar check found in Microsoft Word is an expert system where most grammatical rules are integrated into the programs through a series of algorithms. Yet, expert systems established those rules based on common sentence structures that are usually kept very short. Thus, when the expert system meets, for example, nested sentences, it has a hard time suggesting a proper correction. One way around this issue has involved the help of machine learning, where the algorithms go through an iteration process to learn based on a very large data set. Even though this may sound like a dreamy alternative, machine learning can introduce a lot of flaws into the program. We have to be careful to vet our data set adequately beforehand.
Coming up with new AI has definitely revolutionized how we do things. Now, most bank security measures include AIs for their fraud detection. This implementation has allowed the banks to identify more frauding attempts and has accelerated their detection rate. Hospitals and clinics use AIs to help them with resource scheduling. For instance, when they need to plan work schedules, budgets or material resource allocation. Engineers will use it to perform complex analyses in signal processing and control theory. Manufacturers prefer using AI for automation, whereas retailers will favour the use of AI to perform customer services. Other fields will employ AI to optimize safety systems or machine efficiency, but whatever is the reason behind their use, we can all agree on one thing; the presence of AI will become ever more widespread.
I thank you infinitely for reading this post and if you would like to know more about the mysteries that surround us, please join my subscription list to keep up with my newest content. If you have any questions, please add them to the comment section and I’ll make sure to answer them as soon as humanly possible.
Get new content delivered directly to your inbox