I am not an engineer, just a scientist, but if I would be an engineer willing to do something new and interesting, one of the directions I would consider would be text analysis.
I think image analysis is becomeing a bit boring from the technology perspective and working on text analysis may be much more interesting in the next years.
Working on text and speech recognition have the same objective in mind, which is analysing information encoded through language. This means we have to be able to map our knowledge as our language is the way to transmit knowledge.
First, we want to do it unilaterally, by synthetising information provided.
There are already succesfull examples of automated information synthesis, for example from scientific articles.
Further challenge will be to have bi-directional communication on the complex topic, which is called conversational AI.
There are already some attempts to do it, such as chat box. But conversational AI is still in its infancy.
Why image analysis becomes boring from a technology perspective?
From my current observations, I have an impression that the current progress in image processing is the matter of two things:
Refining algorithms to make them faster and less energy consuming.
The core approach is not really changing. Which makes perfect sense, because I think once some new idea is developed – which is a step change, like deep learning, later it becomes the matter of refinement.
Finding the right applications.
The good example is healthcare. Let’s take medical imaging, which I know a bit from my own experience. Everyone is searching for imaging biomarkers, but the main challenge here is the understanding of the biological problems, to pick up the right data and having access to them, not the data analysis itself. This is of course taking into account the current tools. It was an increadible work to arrive to the point where we do not need to make every computation with pencil and paper.
This however does not mean that the field of AI-based image analysis by itself becomes boring!
I only think, that the most interesting part of this work will not be in software development, but on the side of specific domain knowledge field. For example, medical doctor who knows how to use AI, can have a very interesting journey in developing new disease biomarkers.
Many standard alghoritms are becoming available as ready to use boxes. This is still work in progress, but I am convinced that in few years it will be possible to do very interesting projects on AI-based image analysis without coding skills and standard alghoritms will be available as no need for coding boxes.
Therefore, I think that, if you are an engineer, you either build tools for automated AI (building blocks to allow others to use AI in a simple way), or you run away from the field! ;)
Why text analysis becomes interesting from a technology perspective?
The new challenge in high tech industry is text analysis.
Why is that?
A lot of AI approaches are inspired by a biological brain, as this is now the only truly intelligent system we know. Biological brain, not just human brain, because I think it is not fair to say that human brain is the only intelligent one :)
Brain processes text and language-based information differently than images. Therefore, perhaps computers also should do so. Algorithms based on linear algebra, which have been so successful for imaging data, seem to have some limitations here, as the signal to analyse seems to be much more complex.
Deep learning had some success in text analysis, for example in translation, but to pass the threshold of the analysis of relatively straightforward information, many believe that a new approach is needed.
From the neuroscience perspective, which is more familiar to me, it also makes sense, as our brain processes language differently than images and this process is much less understood.
This is why I think there is a lot to be discovered in the next years on how we can map knowledge, so that we can process complex text (and later speech) information, which could revolutionise many fields, the same as deep learning did.
This is very exciting and I am looking forward to observe this process.
Comments