The topic of artificial intelligence is rather complex, not only looking at the code behind this innovation but also looking at the distinct areas related to it. Apart from the complexity of the name itself, it is also important to know the nuances between each of its disciplines as we dive deeper into what is now considered the fifth industrial revolution, the AI Revolution.
My team at RGF Professional Recruitment is responsible for this new area. We recognize the need to continuously and actively obtain a broader understanding of technical topics and issues relevant to this ever-evolving subject. This allows us to hold meaningful conversations with both our clients and candidates, as well as possess precise information that’s necessary to match a candidate to the right company and job.
Since job descriptions do not typically cover every aspect needed to judge the qualification of a candidate, my team ensures that a thorough discussion is held with both the client and candidates to determine the technical skills and soft skills necessary so we could make a projection of a candidates’ success at the job. With advanced technical expertise on our specialization, all our teams are equipped to make excellent recommendations on both sides. For me, gathering this knowledge is especially exciting in an area as new and innovative as Artificial Intelligence.
Strong AI vs. Weak AI
Today, Artificial Intelligence is defined as “a branch of computer science dealing with the simulation of intelligent behavior in computers” and “the capability of a machine to imitate intelligent human behavior”. AI itself can further be differentiated into “Strong AI”, “Weak AI”, and “Explainable AI” (XAI), depending on the objective to be achieved. As such, strong AI is an “area of AI development that is working toward the goal of making AI systems that are as useful and skilled as the human mind”, which means they can think and perform tasks on their own without input from external forces. Conversely, its counterpart, weak AI, will only focus on one “narrow” or specific task. A good example would be Apple’s Siri or a Poker AI, which has all the rules and moves coded and fed into its system; a weak AI can only deal with data that it has been trained for and therefore is not able to develop new concepts on its own.
Looking at AIs that can beat grandmasters of Chess and Go, one could be led to believe that Google DeepMind’s AI, AlphaZero, is a good example of strong AI. It can play Go, Chess, and Shogi as well as replicate and go beyond the means by which humans play the game. In fact, it often makes sacrifices to utilize positional advantages that human players would never possibly think of. However, in spite of its advanced capabilities, the games it covers are still fairly “narrow” in scope, i.e. two-player grid-based board games. The real world is far more complex and even AI as seemingly complex as one that can play strategic video games like DotA is still beyond our grasp at this time. Therefore, although quite impressive, AlphaZero cannot still be categorized as strong AI.
Explainable AI (XAI)
Another important characteristic of AI would be explainable AI. XAI is a “sub-category of AI where the decisions made by the model can be interpreted by humans, as opposed to “black box” models” in which the decision-making process cannot (easily) be explained by neither the computer nor the researcher behind it.
One example of a simple XAI model is the decision tree. The potential decisions, as well as their potential outcomes, are noted down in a tree-like model, where each branch leads to a conclusion.
For example, you want to decide whether or not to apply to a certain post-graduate institution. The first variable can be “Does it offer the classes I want to take?” from which two branches (“yes” and “no”) emerge. “No” leads to “Disregard”, while “yes” takes you to another decision. You need to check whether you are qualified to match the admission requirements. Again, two branches emerge (“yes” and “no”), of which only the “yes” branch finally leads to the decision to “Apply”.
An explainable AI is able to build much more complex decision trees, of course.
After all “[The] ability to explain the rationale behind one’s decisions to other people is an important aspect of human intelligence,” so having an AI explain how it got to its decision, has many benefits for us as well, including verification of the process, improving the system, learning something new, and discovering things and connections we have never even thought of before the use of AI.
In an interview, Soumitra Dutta stresses the importance of AI and its interaction with people. She addresses the importance of driving people away from “fear” towards AI in a more hopeful and positive direction. After all, AI is supposed to empower humans, assist them in decision making, and people need to feel that their abilities are being complemented instead of being questioned or undercut. In that sense, it is equally as important to retain human judgment while utilizing AI in certain areas to avoid bias and ensure fairness.
Bias in AI
As a last point, it is important to address the topic of bias in artificial intelligence. An article by Jake Silberg and James Manyika of McKinsey displays the opportunities to either use AI to identify and reduce the effects of human biases, while the other opportunity is to improve AI itself to prevent them from creating biases on their own.
Subjectivity is a matter that will shape the decision-making process for humans. In fact, these are some factors that lead us to having unconscious bias:
Although structural decision making is ideal, some people lack the necessary information to make the right decision and end up with arbitrary choices at times
Some people have a tendency to create a different version of the truth
Some people don’t completely understand which factors influence their opinion and how these factors operate to impact their decisions
AI algorithms, on the other hand, can reduce that bias, either through improving decision-making itself or reducing the subjective interpretation of data. More importantly, AI decisions can be questioned and re-examined.
However, studies have shown that the data used to train the AI can be a direct source for AI models to exhibit biases and consequently, deploy these judgments during their work. Oversampling, certain word embedding, or data that already contains human decisions could all lead to bias in artificial intelligence.
The problem in tackling bias (be it in humans or artificial intelligence) is the individual’s notion of “fairness”, according to the McKinsey article. Is it fairer to set different decision thresholds for different groups? Or is maintaining a single universal threshold the right way to be perfectly fair? Currently, “a model cannot conform to more than a few group fairness metrics at the same time”, so the question of when we will achieve perfectly fair models remains.
The article continues to say that to support AI in a fair decision-making process, it might be necessary to pre-process data to maintain accuracy, or, contrastingly, post-process the predictions made by the model in an effort to satisfy fairness constraints.
On another note, researchers have been quite successful in text classification tasks by adding more data points, while facial recognition could benefit from transfer learning. Another characteristic that can help achieve fairness in AI is the above-mentioned XAI which could help identify whether certain factors reflect a bias and could lead to more accountability.
In the end, neither definitions nor statistical measures will be useful in considering the nuances of social contexts at this time. To make sure an AI-supported decision is fair, human judgment still plays a key role. At RGF Professional Recruitment, we match the best candidates with the best available open positions so we could positively impact this evolving technology.
All of our specialized teams constantly strive to learn more about their fields to make sure they have the knowledge and information needed to excel in their respective areas. As for Robotics, AI, and Data (which includes IoT and Machine Learning), the adventure of new discoveries will most likely accelerate in the years to come.
Take this exciting journey with us and find out what kind of open positions we support today!
If you're interested to learn more about AI, watch out for our upcoming blog posts, which will further discuss chatbots, deep learning, machine learning, and many more.