Myths and misconceptions about AI.
Around artificial intelligence (AI) a certain mystical halo has been created, largely caused by the great ignorance that exists in this field, especially in the media. We have all read phrases such as: “with AI, most jobs will be automated and people will be unemployed” or, “robots will take over the world”, or, “it is a science fiction technology”, etc.
As we see, ignorance takes us to extremes, there are those who think that AI will dominate the world and others who think that it is nothing more than a buzzword. We believe that the truth is somewhere in the middle. And, in fact, we believe that only if we fully understand how artificial intelligence (AI) works and where its limitations lie, will it be useful to us.
Only if we fully understand how artificial intelligence (AI) works and where its limitations lie, will it help us.
That is why, below, we will try to dispel some common myths and errors that occur in this field.
Myth 1: AI (artificial intelligence), ML (machine learning), DL (deep learning), different words mean the same
AI is an umbrella term that includes a broad set of computer engineering techniques, ranging from ML and rule-based systems to natural language processing and optimization techniques.
Within AI, there is a subset called machine learning (ML), which is defined as the field of study that allows machines to learn without being explicitly programmed. Machines learn through a process called “training” thanks to the information we provide them.
Finally, deep learning techniques are a type of machine learning (ML) that provides amazing advances. Their techniques are based on the use of what are called artificial neural networks.
Myth 2: AI can only replace repetitive jobs
AI offers solutions to obviously replace repetitive tasks, but it goes further and allows companies to make more precise decisions using predictions, rankings, and groupings that help you solve complex problems.
Some examples where AI is currently used would be:
- Write articles
- Determine the price of hotel rooms and airplane seats
- Advice on wealth management (“roboadvisors”)
- Determine when and for how much insurance claims must be paid
- Determine which ads to show to Internet users
- Execute currency transactions
Myth 3: AI is unbiased
AI technologies are based on data, rules, and other input from non-human data scientists and software developers, and as such are inherently biased in one way or another, with that bias just passing through. to technology.
At the moment, there is no way to completely eliminate prejudices; however, we have to do everything we can to keep it to a minimum. Although the bias cannot be eliminated 100%, we can try to reduce it.
Some recommendations would be:
- Frequently evaluate the data sets that are used for training and validation in ML.
- Selection/confirmation bias can be reduced by creating a team of AI experts by having them review each other’s work.
Myth 4: my company doesn’t need to use AI
Most companies should investigate how this technology can be applied to a wide variety of problems and consider the potential impact of applied AI on their medium to long-term strategy.
Even if AI is not an immediate solution to their problems, organizations should periodically review the decision not to implement AI, since in many ways this option could put the company at a clear competitive disadvantage.
Myth 5: AI learns by itself
The approaches of machine learning can usually automatically determine the necessary parameters. The program or machine learning model obtained is the result of a process of optimization of mathematical parameters, which is mainly executed autonomously. However, before this happens, experienced data scientists narrow down the problem, prepare the data, determine the appropriate data sets, remove potential bias in the training data, and most importantly, continually update the software. to enable the integration of new knowledge and data into the next learning cycle.
Currently, most of these tasks cannot be automated. Experienced personnel is needed to perform these tasks. The entire loop must be run manually (again) to allow for the integration of new insights and data in the next iteration of the model. In the European study referred to above, the lack of qualified staff was the main concern of all organizations.