Future of AI: No data, No Problem
When we started building our services business, Hornbill Labs, Inc, we were startled to see that in spite of such fanfare, medium-sized firms had little or no AI capabilities. As we progressed, it dawned on us that the primary reason for the lack of adoption was the lack of data.
Even now, most firms don’t see their daily business data as a source of insights, beyond excel charts. This lack of foresight prevents them from collecting more data going forward. As firms grow and mature there comes a point when they realize that it’s not possible to compete with industry juggernauts without predictive analytics and ML-based forecasting. By this time it’s too late and the damage has already been done.
Large neural networks and ML models that require mountains of data are not of use to SMBs and early-stage firms. Further, they miss out on a number of edge cases. Take the example of iPhone X’s facial recognition in the early days, although it is trained for thousands of hours on millions of images, it fails to recognize “morning faces” -- the puffy, haggard look on first awakening. This is mainly because it was trained on normal photos of people during the day, and there is less data for the mentioned situation.
Most adopters of Hornbill.AI are early-stage startups or mid-sized firms. Although we hope to be there with them from the very beginning, more often than not, Hornbill.AI has to deal with broken, fragmented datasets that are not ML-ready. We have been developing techniques to work efficiently with such data. Under the hood, we are experimenting and improving.
Hornbill.AI has a very robust automatic data augmentation workflow that adds relevant data from a variety of sources and helps make fragmented data, machine-ready. Recently, we experimented with partial models that work on very limited data. The results were exciting and we will start beta testing them soon.
The future of AI lies in making the revolution more equitable. This requires machines to learn fast and learn from fewer data points. General-purpose AI is the future and it must be less artificial and more intelligent.