Personally speaking, I left my own job earlier this year due in part to deep disagreements about how my previous company was utilizing and marketing itself as AI-based, even changing its name to include "AI".
But when I left fintech to work in the healthcare industry (totally different role), people started coming to me to pitch their products/solutions. I started getting pitched AI products (harness the power of AI & predictive analytics!), and the use of buzzwords was pretty outrageous. I started wondering...do these companies have any real customers? Who are buying these products, and are they seeing a sufficient-enough ROI to justify these purchases?
Alexa would be considered advanced AI in the 90s, wouldn't it? Google Search or Facebook ad targeting might be, too. They certainly fit some definition of ML.
It sounds like you're thinking of companies that are doing simple data analysis and calling it AI, right?
But I can assure you that customers do appreciate the AI component, especially when explained to them. I do a LOT of education in commercial meetings and demos. Some of our customers we invited to our offices for a more in-depth explanation of our technology.
My experience is that customers' interest is piqued by buzzwords, but if you are able to explain the tech in plain words, they are astonished and the sale is much easier.
To summarize, if you are raising money it may be worth it to buzzword the hell of your deck, but when selling the product the key is to focus on benefits + education.
Disclaimer: Am an AI startup founder (https://optimusprice.ai)
But, it was actually tough to sell, because of several factors. 1) The internal risk team did not really like a technology that they could not understand. 2) It was hard to give an explanation of why it would work to the executives.
In the end, we were getting 2-4 big clients per year.
While there are a few horizontal ML/AI/data science vendors, most AI is invisible, a component in some vertical solution, and consumed by people and businesses that don't really care how the results are produced, as long as it works.
Very few companies are good at applying AI across the board, like we see at Google, but the number of businesses using predictive models at least in some capacity is growing.
Skymind is a machine-learning operations company (one of several): we help businesses train, deploy, monitor and update AI models. The same software is used by telecoms, finance, e-commerce and automotive companies -- horizontal.
There are countless vertical specific companies: e.g. Merlon applies ML to anti-money laundering.
It's just one of many features that's best done with AI. I like it that way. Pitching utility is so much better than pitching how we built something with blockchain, AI and IOT combined ;)
In reality, a true AI/ML company is basically just using AI/ML to create services/solutions for existing issues.
Case in point: google is using ai/ml to improve their speech to text and text to speech services.
DL integration into or as a replacement to pieces of software stacks, plus long term model updates, measurements, testing and avoiding regressions (very hard, think of ensuring all past samples are similarly predicted as before after a model update, that's beyond simply improving the accuracy) is difficult and standard practice is building up slowly.
I can’t speak for them specifically or any other AI company but from cursory chats with folk in the industry suggests a lot of work is consultancy and proof of concept type stuff with bigger companies which they use to partially fund product dev on their own internal projects - some of which see the light of day.
Apart from that, I'm also personally responsible for implementing DL for data anomaly detection.
So yes, ML/DL is very much over-hyped right now, but it has real world tangible benefits. It is being used in multiple fields and I believe will grow quite fast.
That's quite an interesting project you have. Unfortunately probably not really relevant to my use case since two skills are necessary:
1. (good) control of the native language (not english) 2. Understanding and experience at the subject field
I don't want to give too many details but the workflow basically consists of taking a poorly labeled entry and expanding/filling out the missing fields. To do this you both need to speak the language and have some experience in the field.
As for the data anomaly detection I don't do any data annotation
Anyways, sorry it's not really relevant to my case but good luck, I think it's a great project that certainly has a lot of potential. Maybe in a future stage you might be interested with partnering up with a platform like coursera to form your workers to do slightly more complex data annotation jobs (like mine).
Regards
We are releasing our API next week. The idea being you can quickly try out the algorithm and see Results and Training time on your own datasets.
Some products like https://www.contractstandards.com/ are beginning to appear. They need document classification, text sentence level classification, entailment, clustering are all necessary for making contract analysis easier for the user.
And these products need good domain experts to identify pain points where ml can help.
In ml as API side, one potentially valuable product suite I have seen is https://aylien.com/text-api/ . Even they are building additional products like news analysis.
But I am little sceptical on whether ml as API can scale in terms of addressing specific task-specific nuances needed for products building upon them.
But there sure are a lot of stories about adversarial bayesian multi-model deep net retention markov cycle consistent custom asic demystified robust perturbations.
The speech recognition and predictive text on my phone is still awful.
Another point is marketing. A ride sharing service like Uber or Lyft could easily market themselves as "using AI/ML to solve last mile transportation". There's reason to believe they have some ML teams that do amazing work to support the ride sharing product. Both above have paying customers.
Usually their website is buzzwordy but they are not necearily too much vocal because they find their customers through direct contact. And they do have customers... theres a real market
In healthcare there are not ao much killer application yet in ai/ml and its true that a lot are running on investor funds, but in my opinion this will change soon.
So sibling comments mentioned that some companies struggle with selling products not technologies. I agree.
Intelligence is merely pattern matching and goal oriented planning, exactly what machine learning is doing today.
Stop worrying and learn to love the bomb.
I'm rather less confident about this than you may be. When I'm doing something that I could plausibly claim requires "intelligence", then it certainly doesn't really feel like I'm just doing pattern matching. But I'm increasingly of the view that our perception of our own mental processes is a poor guide to their reality. The likelihood that my intelligence really is just large scale pattern matching, even though it doesn't subjectively feel like that, seems quite high to me.
All you've done is taken what machines do today and claimed that is what intelligence is.