As years go by, the technology-related expectations grow. There is this imminence floating about. So many revolutionary technologies have been prototyped and announced, yet so many are work in progress. In this context, let’s meet reinforcement learning.
Reinforcement learning belongs to machine learning. Artificial Intelligence domain, that is. Its inspiration relies in behaviorist psychology. The applications are numerous, from simulation-based optimization to swarm intelligence.
Reinforcement learning in simpler terms
Software or, if you like, machines endowed with intelligent software, learn new behaviors based on feedback. The environment sends feedback and guides the AI towards the ideal behavior.
The main application of reinforcement learning would be a viable decision-making process in what machines are concerned. Think of driverless cars and their intelligent software, for example. The online media covered the debates on the way such software can make choices and act upon sound decisions when people’s life is at stake. In such cases AI needs a way to enable machines to behave in the most suited way.
Of course, machines cannot receive rewards in the form of a bright tin foil star or a cookie. However, there are ways of communicating a positive feedback when the AI performs operations in the desired way.
Good decision sequences are of the utmost necessity when leaving certain activities to robotic hands, so to speak. That is why developing reinforcement learning strives to deliver in 1 up to 2 years in the future. The goal here is to make computers figure out solutions that programmers cannot come up with. These solutions would allow for even more developed computers to come into action.
Deep Mind and reinforcement learning
The Deep Mind website explains RL in terms of learning via trial-and-error. For that purpose, the Deep Mind team employed Atari games. Their Deep Q-Networks (DQN) algorithm played games on the Atari 2600 console, and received reward signals for high game scores.
The experiments illustrate the way the DQN managed to achieve human-like results. As the team puts it , the experiment was a success and the AI displayed a â€œremarkable progress on a wide variety of challenging tasks â€œ.
Of course, Deep Mind does not hold the monopoly in RL research, Various experiments take place all over the world, and it seems they also have satisfactory results. This AI training method is apparently showing tremendous progress. Here is another instance of algorithms evolving via RL.
Artificial Intelligence â€“ on the cusp of becoming common
There already are a lot of applications, software programs and gadgets that employ AI. Due to its numerous emerging points, we should soon forget that AI was a novelty long ago. Nevertheless, as it is the case with humans, not all machines show an equally developed intelligence. Some algorithms are more modest, others are the state of the art.
If we talk numbers, and not performance, AI is indeed on the cusp of becoming common. Due to its (yet) high costs, it may not have crossed into commercial convenience yet. Nevertheless, as it was the case with computers, how many of us, when in the presence of modern AI, take a moment to think about it? Especially when it comes in code form. Revolutionary design established a link between a futuristic device and high-end AI, but invisible algorithms cross straight into public subconsciousness.
Reinforcement learning speeds up the rate machines learn. In a matter of hours, machines can learn to perform new moves and consequently to execute new tasks. Here’s something to think about.
Reinforcement learning, AI and businesses
What would the link be here, one may ask. Besides the obvious link represented by the companies specializing in all AI-related operations, once AI will take off, a major disruption is to be expected. From job switching to automated employees to different marketing interfaces, full throttle AI could use some strategic anticipation.
The short list of RL practical applications varies from one respondent to another. Still, we can note that employing RL-trained machines in economics and advertising do appear among the candidates.
Although there are voices that point out that machines don’t inherently employ ethics in approaching any given task, the progress hardly takes this side into consideration. Let’s elaborate. Researchers have analyzed the effects of AI decisions when put into practice in the labor market or in restructuring institutions. The machines have issued the optimal number of employees to be let go in order for the company to redress itself. Yet nor did they know or taken into account that not all employees are inter-replaceable. The results proved that AI cannot give sound advice when it comes to letting employees go, since it cannot compute all the possible outcomes.
Therefore, while in some aspects businesses can rely on algorithms to support their decisions, in others the traditional approach would be better. For now, at least.