Dealing with AI / ML - Insights and Challenges

Artificial Intelligence, Machine Learning & Deep Learning (AI, ML & DL) are being increasingly looked at w.r.t bringing in benefits of automation and reducing human limitations or bias in the system. I recently attended some online sessions on AI, ML & DL, where presenters shared some good perspectives. I learnt that, while AI, ML & DL can definitely be used and are already being used effectively in certain areas, as professionals in this Fintech & Digital space, we also need to be mindful of certain aspects while dealing with AI, ML & DL. I list these learnings as pointers below: -

a. The assumptions we make on the AI, ML & DL models are very important. So, this is more managerial than just technical.

b. Data set and training of model needs to be an ongoing loop. Depending on use case & learnings, the frequency needs to be regularly decided.

c. The models themselves can be biased & tend towards an opinion. So machines are not as objective as we feel.

d. Any model you create, you need to also let go with some acceptable cases of exceptions. Also, the more accuracy you seek the complex is the model to create.

e. Creating models also poses a lot of ethical questions w.r.t identifying differences, categorizing/ segmenting so need to handle delicately.

f. Given biases, ethical and privacy issues involved, there are many regulatory challenges w.r.t adoption of AI, ML & DL.

g. While models are evolving, the challenge is still with Data. (this is something I identified in my recent book 3F: Future Fintech Framework too)

h. In era of personalization, clustering, segmenting or even micro segmenting will lead to one size. Remember "One Size Fits no one".

i. Anonymizing data helps and must be done, but at aggregate level biases can still be present.

Happy to hear from experts on the same.

Besides the above, I also read a Dec 2020 report by Trend Micro, titled "Malicious Uses and Abuses of Artificial Intelligence". It offers some insights on various aspects where AI can be used in wrong ways. Some key areas of concern are as below: -

- Stock Market Manipulation
- Social Media Impersonation & Social Engineering
- Captha Breaking
- Escaping, Abusing Image & Voice Recognition
- Escaping Spam Filter
- Password guessing & Hacking
- Online Game cheats
- Content Generation & Manipulation
- Robo Calling
- Deepfakes