Photo from Unsplash
Companies are lying a lot. Not just small startups trying to get funding but even big companies trying to maintain their share price. It is said that companies which mention the word ‘AI’ in their annual reports are seen with a higher positive outlook. I read the below tweet and a flood of stories came to me.
[ This piece is highly opinionated. ]
Some real stories about fake AI orgs
AI based recruiting
What actually happens - Some Elasticsearch; some handwritten rules to match candidates to CV
AI based calling
What actually happens - SQL query to find the best time to call
AI based customer support
What actually happens - People answering the chats most of the time
AI based SQL query parsing
What actually happens - Run through a lot of rules. If it fails, get it done by humans.
AI based speech to text
What actually happens - Use commercial speech to text APIs
AI based chart/report generation
What actually happens - Use commercial speech to text APIs. Parse the detected entities through a SQL template.
We use RL for active learning
What actually happens - Faster periodic retraining of supervised models
What’s the main question?
There are 2 questions to ponder upon
Does a company which use commercial APIs to develop AI products be called an AI company?
Does a company which use opensource AI models out-of-the-box to give AI solutions be called AI companies?
What is not an AI org (IMO) ?
To make AI means to make something that learns from data.
I strongly feel that in both the above cases, the company cannot write we do AI for x or we are AI based x. They are not AI first companies. Using external API doesn’t require AI engineers and working with open source models requires just another regular Python developer. In India, you can get such an engineer at just $5/hour.
What is an AI org (IMO) ?
A company with models trained with its own data and in production
A company with AI research publication but nothing in production
A company which makes products used by others to make AI of their AI products
Although Transfer learning is not anything niche or cool anymore. But let’s give these companies some credit for collecting data & MLOps. Albeit not many orgs know proper MLOps also.
Now how to detect fake AI orgs?
Indicators of fake AI orgs
Companies which do not have any data scientist with past experience of AI is mostly lying.
Many companies hire regular python developers and give them inflated titles like Data scientist, AI engineer or ML engineer to make it look cool. This is also done intentionally sometimes by service-based companies to look more credible to get clients.
No publication/blog definitely makes things look doubtful. Having publication clears the doubt.
Companies which do open source for AI are the strongest.
Founders with only business background make some of the biggest liers. Never trust them easily. They do x and say y because they themselves don’t understand AI.
How does this affect us?
As a data scientist, I don’t want to interview or end up at wrong places.
As an investor, I don’t want to invest in fake companies.
Some edge cases to ponder upon 🤔
Is a company using a linear/logistic regression an AI company? Do we consider classical ML company as AI company?
Is a company using AutoML an AI company? It doesn’t require much brain to use AutoML.
What about those which use AWS, Azure or Google AutoML - basically no Modelling or MLOps required. Your job is just to collect the data.
Would you join such companies?
Come join Maxpool - A Data Science community to discuss real ML problems!