How close will 2020 bring us to perfect computer vision and AI?

We believe that in 2020 AI will continue to evolve and provide new challenging opportunities for entrepreneurs and enterprises

The previous year of 2019 was notable in terms of the AI and computer vision technologies integration into the usual lives. We believe that in 2020 it will continue to evolve and provide new challenging opportunities for entrepreneurs and enterprises. Let’s figure out the upcoming trends and try to analyze their future impact.

• Improvement of dialog capability and smarter context handling in AI assistants

Until now, AI products have been quite limited in terms of audial or text dialog with humans. Yet in 2020 a new trend could appear, with proper dialog and a better understanding of context significantly increasing the quality of personal assistants (Siri, Alexa, Google), as well as enhancing the experience of using them. Predictive text applications will use context more to better understand what you want. When coupled with voice recognition, input for prediction will increasingly include indications of your state or mood to also better guess what you want, while understanding the specific context.

• AI will work with smaller datasets

Currently, big data is simply not available or is expensive for many real-world needs, or it is just a collection of non-attributed parameters, which makes its usage very complicated. There are two ways of tackling this problem. Firstly, you can try to acquire additional data by yourself. Secondly, you can solve the issue of the performance of deep learning on small datasets. The first path could be completely impossible or very costly, for example, in such a demanding domain as healthcare, because the development of global medical data collection and storage standards could be a task for years. Luckily, recent approaches to deep learning with small datasets and pre-training can now give reasonable results for less time and money, which opens a huge market in different necessary domains. So, we can expect considerable potential over the coming year for startups and enterprises that can develop effective solutions based on small datasets for specific markets with very expensive data.

• Eliminating bias

One of the problems with deep learning and small datasets discussed above has been the susceptibility to bias. For example, small datasets of people may not include all ethnic groups. This can make it impossible for an AI program to recognize or behave properly when confronted with these ethnic groups afterwards. More generally, AI (and its developers) will need to make progress with integrating ethnic studies and eliminating bias. Algorithms will need to be taught to behave in a smarter, yet more natural way for human understanding.

• Regulation and ethics

The core challenge AI is facing now is the issue of privacy. Many users are sharing their doubts and skepticism about face recognition, mass personal and even medical data collection and more. There is no doubt that governments clearly see that AI will soon be even more deeply integrated into the daily needs of common people, so they are already creating regulations for organizations developing and utilizing AI. For example, European GDPR and Californian CCPA were created in order to control personal data collection and usage; these regulations are controlling potential data to be used in AI systems. Also, starting in 2020 many more countries will make efforts to regulate AI. The major trend here is finding new and secure business models of AI usage that protect users’ privacy, which in turn will result in natural adoption, increasing the trust of such services in citizens.

To sum up, although our understanding of artificial intelligence was quite clear in 2019, end users are asking more and more questions, raising concerns about data security and ethics. As AI continues to move along its current trajectory, it will likely become a technology that “thinks” more like humans, which will shift understanding in humans and increase trust in the technologies. It means we will know enough to get significant business benefit out of AI for the foreseeable future.