The emerging technologies like Artificial Intelligence and Machine Learning have huge potential and offer great benefits to the industry. More and more companies are deploying these technologies to boost their growth further. The new evolving technology trends are expected to take the industry’s growth on next level. Commenting about the upcoming trends, Satish Pala, Chief Technology Officer, Indium Software shared below trends for 2022.
AI is increasingly being used for automating core but mundane jobs like auto-KYC as-well-as for achieving novel goals like assessing sleep health w/o a doctor. A lot of ongoing research has led to development of very advanced algorithms. The prevalent open source culture has helped in democratizing these algorithms in the sense that even a junior Data Scientist can implement advanced algorithms as a Github repo or a blog is always available. Having said that, the core challenge for ML remains the availability of quality annotated data. Many companies, especially startups who envision AI as core component of their product, struggle to collect enough data to train a model which can make accurate predictions. As AI pioneer Andrew Ng says, ML Development needs to move from being model/algorithm-centric to data-centric. We hope that the hype around ML/AI is met with the basic understanding that nothing useful in ML can be built if you don’t sizable and heterogenous and well annotated training data.
The security concerns around AI are more from the perspective of the bias it can create in the decision making. AI models learn the patterns in the training data and generalize it across the population. So the biases of the training data would creep in in the model. For example, if a text generation model is trained on text corpus which addresses common gender professions like doctor, teacher etc. with a male pronoun, the text generator would also use male pronouns and carry the bias against female pronouns. Similarly, if certain employees of a bank deny loans to a particular community, the model would also learn to discriminate against that community in giving loans.
Another concern is around privacy especially when Personally Identifiable Information is present. To address this, Differentially Private algorithms and methods are used which create a tradeoff b/w privacy and accuracy. So basically some (systemic) noise is added to the data so that the data can’t be traced to the person who it belongs to so the privacy is ensured. But the model accuracy takes a hit because of introduction of noise. But this hit is tolerable and all the big tech companies like Apple, Google are already using these algorithms in their devices.