One of the drawbacks to the current revival and popularity of Artificial Intelligence (AI) is that we see a lot of vendors, professional services companies, and end-users jumping on the AI bandwagon labeling their technologies, products, service offerings, and jobs as AI goods, jobs, or offerings without always being the case.
On the other hand, there isn’t a well-accepted delineation between what AI is definitely and what’s certainly not AI. That is because there is not a well-accepted and regular definition of what is artificial intelligence. Really, there is not a normal definition of intelligence, period.
Maybe it is ideal to start with the general aims of what we’re trying to achieve with AI, rather than definitions of what AI is or isn’t. Since the start of the AI in the 1950s, the goals of smart systems are the ones that mimic human cognitive skills.
This means the capacity to comprehend and understand its environment, learn from training and its experiences, make conclusions based on reasoning and thought processes, and the development of “instinct” in situations which are vague and imprecise; essentially the planet where we reside in. After all, AGI systems are attempting to create systems which have all the cognitive capabilities of people, and then some. So certainly all AGI initiatives as AI initiatives.
On the reverse side, simply automating things does not make them intelligent. It could take time and effort to train a computer to understand the difference between an image of a cat and a picture of a horse or even between different species of dogs, but it doesn’t indicate that the system can understand exactly what it is considering, learn from its experiences, and make conclusions based on that understanding. , but it does not mean that the assistant understands what you are in fact talking about or the significance of your words. So, can we really argue that these systems are smart?
In fact, he argues, most of what’s being branded as AI in the current market and media is not AI in any respect, but rather just different variations of ML where the programs have been trained to perform a specific, narrow task, using different approaches to ML, of which Deep Learning is currently the hottest. He asserts that if you’re trying to have a computer to recognize a picture simply feed it enough data and together with the magic of math, statistics, and neural nets that consider different connections less or more over time, you’re going to find the outcomes that you would expect. However, what you’re doing is utilizing the human’s understanding of what the picture is to make a large data set that can then be mathematically matched against inputs to verify what the human knows.
How Can Machine Learning relate to AI?
The view espoused by Professor Perez-Breva is not isolated or outlandish. In fact, if you dig deeper into these arguments, it is hard to argue that the narrower the ML job, the AI it actually is. But does that mean that ML does not play a role at all in AI?
The line between intellect and just automation or mathematics is a tricky one. Should you decompose any smart system, the eventual end goal of AGI, then it is going to look like bits and bytes, neural networks, decision-trees, lots of information, and mathematical calculations. In the same way, if you decompose your brain, it’s only a lot of neurons firing electrochemical pathways.
Are individuals intelligent? Are zebras intelligent? Is bacteria intelligent? Where is the delineation between intelligence in living organisms? Maybe intelligence is not truly a well-defined thing, but instead monitoring of the characteristics of a system that exhibits certain behaviors.
In this light, one of these behaviors is understanding and perceiving its environment, and another of those is learning from experiences and making conclusions based on these experiences. In this light, ML definitely forms a part of what is vital to earning AI work.
Over the last 60+ years, there are many approaches and attempts to acquire systems to learn how to understand its surroundings and learn from their own experiences. These strategies have included decision trees, association rules, artificial neural networks of which Deep Learning is one such strategy, inductive logic, support vector machines, clustering, similarity and metric learning such as nearest-neighbor approaches, social networks, reinforcement learning, genetic algorithms, and associated evolutionary computing procedures, rules-based machine learning, learning classifier systems, sparse dictionary approaches, and more.
For the layperson, we would like to stress that AI is not interchangeable for ML and surely ML is not synonymous with Deep Learning. However, ML supports the aims of AI, and Deep Learning is one way to perform certain aspects of ML. Or put it another way doing machine learning is necessary, but not enough, to attain the aims of AI, and Deep Learning is an approach to doing ML which may not be adequate for all ML needs.
What Parts of AI are not Machine Learning?
It is an interesting exercise to consider how you, as an adult individual, have obtained the intelligence you have now. On occasion, you learned from simply being part of your environment like studying how gravity works, how to talk to other people and understand what they are saying, and cultural standards.
In other cases, you heard in a teaching environment from instructors who knew a specific abstract subject field such as math or physics. In yet other cases you learned from repeating a particular task over and over again to get better in that task, such as sports or music. From the AI standpoint, these are only different kinds of learning, and therefore, different machine learning approaches.
Supervised learning for being taught how to do things. Unsupervised learning when you are learning from observing the world. Reinforcement learning when you’re learning through trial and error. Therefore, does not it make sense that all kinds of machine learning ought to be considered AI? What else can there be?
Some say this machine learning is a sort of pattern recognition, knowing when a specific pattern occurs in nature or expertise or through sensations and then acting on that pattern recognition. When you look at it from that perspective, it will become evident that the learning part has to be paired with an active component. Decisions and reasoning aren’t only applying the same reaction to exactly the very same patterns over and over again. If this was the case, then all we’re doing is using ML to simply automate much better.
Given the same inputs and feedback, the robot will execute the same action. However, do people really work like that? We experiment with different outcomes. We weigh options. We respond differently when we’re worried than when we’re relaxed. We prioritize. We think ahead and think about the possible outcomes of a decision. We perform politics and we do not necessarily say what we would like to say. We have self-consciousness. We have “awareness”. These are the frontiers of AI.
The Moving Threshold of Intelligence
In reading this bit, you are actually thinking and studying about Machine Learning and AI, the connections to each other, and whether or not specific ML actions are accomplishing the goals of that which we aim to achieve in AI. Similarly, even for those at the extremes of the AI spectrum considering just AGI to be genuinely AI or about the other polar opposite that considers any use of ML to become AI, the reality lies somewhere in the middle.
Some machine learning projects are more like automation and application of formulas that can’t continuously evolve or react to change, while other machine learning attempts are somewhat closer to the intellect, which can change and adapt more with experience, improving their job or desired outcome.
The tech industry continues to iterate on ML and address issues previously considered to be more complex and hard. Since the selection of ML activities older, while others are definitely not AI-like whatsoever or particularly intelligent, others are progressing the industry down the route of AI. Eventually, we’ll begin to see the sort of technology evolution that has long been the aim of AI.