The machines are slowly learning but they can and will be fooled

We get over 400 distinct bots that scan our website each month, the large majority of them look at the sentiment of our posts to provide signals to trading systems (well at least they claim to do that). Is machine learning helping shape the future? Is Artificial Intelligence going to drive the financial world?

The answer to both those questions is yes and no. There are several examples where Artificial Intelligence is being used in the financial world,

1. Image processing is being using to scan and clear or cash cheques/checks in real time.
2. Car driving pattern boxes or telematic devices which track how a driver is driving are helping drive down (or up) car insurance premiums.
3. It is claimed that data from wearable devices is being used to provide better health insurance premiums.
4. Companies are constantly studying customer feedback and sentiment from social media.
5. News articles and opinion pieces (such as ours) are being scanned for market sentiment.
6. Machine learning is being used to process millions of pages of legal contracts and automatically draft them. JP Morgan claims to have mastered this. It claims that software it built does the work in seconds what took Lawyers 360,000 hours.
7. Biometrics such as voice, finger prints, and iris eye scans are increasingly being used for authentication and authorization of transactions.
8. Billions of hours of audio and video are being analysed to fight financial crime.
9. Content analysis is being deployed to understand regulatory orders and general notifications.
10. Satellite images from car parking lots is being analysed to understand mall usage.
11. Review websites are being analysed to predict brand trends and financial performance.

All of the above sounds great, doesn’t it? But is all or any of the above accurate?

We will start in reverse order of the list. Review websites are increasingly biased and can easily be manipulated. How else did a Garden shed become the top-rated London restaurant on TripAdvisor after the site was tricked by fake reviews. And most news is overwhelmingly negative, yet we don’t see algos sell, quite the reverse.

And on reviews – Most of the lowest rated companies end up being the best financially and the ones with great reviews end up running into financial troubles. The worst rated Airlines in Europe on TripAdvisor are the most profitable, the best rated ones all went bankrupt.

It is claimed that data from car parking lots has predicted trouble for a number of retailers and malls. But does the presence of cars automatically imply people are actually shopping? A study recently proved that studying those images was no better than randomly predicting how malls are doing.

Content analysis sounds great, the only issue is structured content is one thing, unstructured content an altogether different game. And we wish machines understood sarcasm.

Analysis of hours of audio and video sounds great but if it was really that great then you wouldn’t need human operators at Facebook, YouTube and the likes to moderate and remove disturbing content.

The use of biometrics for authentication and authorization should effectively lower fraud rates. But it hasn’t. Fraud rates on financial transactions is increasing, not decreasing.

Replacing lawyers with machines is brilliant, that explains why law firms have never been more profitable. Or the fact that law firms are now paying $200,000 for junior solicitors.

Scanning news articles for sentiment, we wonder what those bots make of our posts? And given that there is barely any positive news out there doesn’t seem to have made an impact on these bots. Everything is awesome in the markets.

Companies studying social media for trends and sentiment – if they really did then shouldn’t customer service improve?

Data from wearable devices is being used to drive down health insurance premiums – if that was indeed happening you would have enough cheat devices by now.

Car driving devices or telematic devices are free. Think about it – the only way insurers can recover the cost of it is through higher premiums, someone must pay more.

Finally cashing cheques with just an image, great in principle and great in practice as well. At least the machines got that right.

The machines are slowly learning but they can and will be fooled, let’s not forget that it is humans coding the machines.

Related: Does the global economy run on fake data?

Leave a Reply

Your e-mail address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.