Artificial intelligence is still a long way off. Big companies like Google are suggesting that we are only living in the first era of artificial intelligence as machines begin to perform speech learning and image recognition. For now, we have smart assistants who still speak like machines, which is far from the technology’s true mission.
For more information and assistance in this environment, it is worth contacting artificial intelligence solution providers.
Artificial intelligence faces three main challenges: the first is related to the volume of data, the second is to limit multiple tasks, and finally to the compression of inferences.
The amount of data used
The need for AI to use data to explore its environment is well known, but the amount of data involved is not regularly accounted for. Artificial intelligence not only requires more data than the human brain to understand concepts and characteristics, it needs thousands of times more data than a human needs.
Neil Lawrence, a professor at the University of Sheffield and a member of the Amazon AI group, noted: “It has been observed that successful results of deep learning artificial intelligence are cases where machines have access to a huge amount of data.”
Successful examples of artificial intelligence include Facebook and Google, companies that have access to massive amounts of data from their followers, users, and the web in general, which allows these brains to search more efficiently and logically and achieve better results.
A sensitive area where AI data collection needs to be strengthened will be healthcare, where the technology is used for artificial vision tasks such as tumor detection in X-ray scans, but where digitized data can be scarce and can produce false negative results or positive points that can lead to fatal errors.
Multitasking artificial intelligence: an unsolved problem
Human intelligence is universal, it can perform many tasks, and this is precisely the failure of the artificial intelligence developed so far, since it is applied to a specific field of knowledge or activity. Once the artificial intelligence brain is trained, it can be very effective for tasks such as recognizing animals or being the ideal contender in a video game, but according to Rya Hudsell, a scientist at Google’s DeepMind division, there is such a thing as an artificial neural network in the world. Capable of performing more than one task at a time.
Until now, it has not been possible to create artificial intelligence systems that can learn new tasks, because once you receive new training, you forget the previous accomplishments, so you have no intellectual memory. A proposal for solving this problem already exists and is called progressive neural networks, which create separate deep learning systems, but which, when combined, can transmit certain bits of information.
At the moment, this is nothing more than a promising method in the beta testing stage, which has been implemented in some robotic manipulators such as cyborgs and has managed to speed up the learning process from a few weeks to a day.
Understanding your findings
It remains to be determined how artificial intelligence comes to some conclusions, since at the moment it is possible to determine only the functioning of neural networks in general, but the reasons why an artificial brain comes to a certain conclusion are unknown.
To illustrate this flaw, an experiment was conducted at the University of Virginia in which researchers created an eye-tracking system for a neural network, and it was shown images of a bedroom, and the artificial intelligence system was asked to indicate what it is, having closed the windows, and instead of looking out the windows, he was looking at the ground.
When he saw the bed, he replied, “The windows are covered with curtains.” The answer was correct, but he never saw a window, and he did not have enough data to come to this conclusion, which remains a mystery.
At the moment, it remains to be expected how these three limitations will be removed, so that artificial intelligence goes beyond performing mechanical and repetitive tasks, becoming a real intelligence with multitasking.