AI Isn't Magical and Won't Help You Reopen Your Business (wsj.com) 42
The coronavirus is helping to erode the hype around artificial intelligence; data scientists get the axe and some 'old-fashioned' solutions work better. From a report: When SharpestMinds, a startup that sells mentoring services to data scientists, surveyed its alumni in April and again in May, it found that 6% of respondents had been affected by furloughs, pay cuts or layoffs. That's a drop on the ocean compared to the enormous layoffs in, say, the restaurant business, but it's notable because these jobs are generally thought to be business-critical roles requiring high-demand specialized skill sets. Uber recently shut down its AI research lab, and Airbnb's layoffs included at least 29 full-time data scientists, according to its directory of those let go.
The pain for data scientists will likely increase as companies rethink how they spend, predicts SharpestMinds founder Edouard Harris. Hiring for such roles has slowed significantly, down by 50% since before the pandemic, he adds. On the other hand, that means there's still demand, though it's diminished. What's happening is not so much a reckoning as a "rationalization" of the application of AI in businesses, says Rajeev Sharma, head of enterprise AI at Pactera Edge, a technology-consulting firm. "[Companies] feel this is a time they can get rid of extra hires or lower performers who are not a good cultural fit," he adds.
The pain for data scientists will likely increase as companies rethink how they spend, predicts SharpestMinds founder Edouard Harris. Hiring for such roles has slowed significantly, down by 50% since before the pandemic, he adds. On the other hand, that means there's still demand, though it's diminished. What's happening is not so much a reckoning as a "rationalization" of the application of AI in businesses, says Rajeev Sharma, head of enterprise AI at Pactera Edge, a technology-consulting firm. "[Companies] feel this is a time they can get rid of extra hires or lower performers who are not a good cultural fit," he adds.
What is a "data scientist"? (Score:4)
Re: (Score:2)
We used to call them statisticians. But now with more science.
Re: What is a "data scientist"? (Score:3)
That's basically all it is. Think back to high school where you put a bunch of data points into your revolutionary ti-82 calculator, and then did a regression to create a linear model, and then you could extrapolate or interpolate to have a good idea of where new data points might land. That's basically what AI does, only using a variety of algorithms beyond regression. The job of the statistician is to figure out which algorithm would make the best choice of algorithm, and be able to make corrections and a
Re: (Score:2)
"However, AI as it is (notice it isn't actually intelligent) isn't going to make humans obsolete any time soon, and for some things it's an overly complex solution to a problem."
AI is about as intelligent as a gold pan or casting a mold. It's a handy way of solving certain classes of problem which human intelligence has found difficult to solve otherwise. Defining a set of cuts and steps to make a CNC machine cut a human face out of a block of stone is quite difficult even for an engineer with a genius leve
Re: (Score:2)
AI is useful for looking for oddball correlations that you might not spot otherwise. This applies specifically to big data, where you collect so much data that it's basically useless for analysis due to volume so you run a computer algorithm over it to see if there are any trends.
As this is strictly a tool, you need a data scientist to make sense of it all as well as expose any underlying biases and weakness in your data or collection or analysis. The law of unintended consequences comes into play here as t
Re: (Score:2)
Basically, combine an old school statistician with a CS major with a machine learning and big data specialization. Take it to a PhD (but an accelerated PhD aka a masters called a PhD, there are a lot of hyper accelerated data science courses).
Re: (Score:2)
But now with more artificial science.
FTFY.
Re:What is a "data scientist"? (Score:4, Informative)
Isn't collecting and analyzing data something that every scientist does?
The proper analysis of data is a science itself.
"Data science" is a broad term. It means some combination of statistics, database administration, and machine learning.
BY FAR the most requested skill of people hiring "data scientists" is knowledge of SQL.
Re: (Score:2)
Statistics, db admin and machine learning....So, not science at all.
Re: (Score:3)
Statistics, db admin and machine learning....So, not science at all.
I suppose it depends on your definition of "science". If you come up with ideas (hypotheses) and then experimentally verify or falsify them, many people would consider that "science". You will have great difficulty building a machine learning system without experimentation.
Re: (Score:3)
Bad statisticians make for bad scientists.
db administration should be obvious. Scientists who mismanagestheir data are bad scientists. The alternative is Excel.
Machine learning is mostly statistics. You're probably just thinking about neural networks, but machine learning is about a lot of other very useful, if super boring, things that aren't NNs. Things like K-means clustering, PCA, linear and logistic regression, classification (like KNN, SVC, MMC, decision trees, etc.) Basically, a whole lot of thi
Re: (Score:1)
It is. The difference is that real scientists also understand science and at least one scientific area. "Data Scientists" are not scientists and have no clue about the real world. They are sort-of the MBAs of data-processing, i.e. bean-counters on steroids.
Re: (Score:2, Troll)
I dislike the term "data scientst" for exactly that reason.
A data scientist seems to be someone who doesn't know enough about statistics to be a statistician, doesn't know enough about a particular science to be a physicist, biologist, etc. and doesn't know enough about about algorithms to be a computer scientist or mathematician. There's also precious little scientific method. So basically it means "analyst."
Typically it means you did at least a Coursera course and can cobble together enough lines of unstr
Re: (Score:3)
Uh oh, I irritated a data scientist!
Re: (Score:3)
A data scientist is someone who knows more computer science than a statistician and more statistics than a computer scientist.
Yip, try KISS & old-fashioned first (Score:4, Interesting)
I've been on a few projects to auto-classify messages or documents, and found a list/table of weighted words and phrases to score on works pretty well if you and a subject matter expert study and tune the results over time based on examining actual results.
You may have have to do some specific programming, like detecting phrase order for "split" pairing, and parsing out numbers, but that's only a few special cases. The majority can be driven by the word/phrase match table alone, which power-users can usually assist with. Also allow negative weights.
And unlike neural nets (NN), you can easily incrementally adjust it by tweaking the words and match weights. With NN you generally have to re-run it against the revised training set. Yes, I know there are exceptions, but those are generally harder to arrange.
It kind of reminds me of the commonly found advice to try a "monolith" solution first before using microservices. There are a lot of improvements to monoliths that can be done to stretch them to the limit. Mine those first. For example, if your shop tends to use the same brand of database, then a shared stored procedure(s) can act a lot like a typical microservice if you need to share a service across apps, and usually easier to set up.
Re: (Score:2)
you and a subject matter expert study and tune the results over time based on examining actual results.
So what you are saying, is that you and a highly-paid expert, working for a few weeks, can do just as well as a GPU-based NN running for an hour and using 5 cents worth of electricity?
Re: (Score:3, Interesting)
Neural nets (NN's) usually need a big and reliable training set. If you have one, sure, run it through a NN engine to get a categorization service going. If you don't, building one would take quite a while. I probably should have pointed that out.
Another issue is the difficulty knowing why the NN fails when it does. This is a weak-point of NN's. Putting a tracer log on the matching list is relatively easy. In fact, it's good practice to store the top 5 or so production matches in a log (auto-flushing old r
Re: (Score:2)
There are a lot of classification algorithms. A NN is usually not the right choice.
Re: (Score:2)
you and a subject matter expert study and tune the results over time based on examining actual results.
So what you are saying, is that you and a highly-paid expert, working for a few weeks, can do just as well as a GPU-based NN running for an hour and using 5 cents worth of electricity?
A NN trained on 3 outliers is going to have output similar to random noise. A human using those same 3 outliers will be able to meaningfully fine-tune their algorithms.
Right, R&D suffers with downturns... (Score:4, Insightful)
this isn't new.. This is just typical MBA management tactics When you are facing a downturn and are looking at your expenses trying to find some to cut, the OBVIOUS solution is to reduce R&D spending. Which, on it's face, makes sense.
Yea, it's only a short term tactic, but dumping the R&D doesn't destroy your ability to deliver product and services to your customers so should your fortunes improve, you can more easily expand your production and make money.
What to AI engineers usually do? What department are they usually in? That would be R&D... Your fate as a Administrative assistant in R&D would be about the same, you are all going to be finding different work.
Re: (Score:3)
What to AI engineers usually do? What department are they usually in? That would be R&D...
Most data scientists that I know work in marketing.
Marketing budgets tend to also be disproportionately cut during downturns.
Re: (Score:2)
Re: Right, R&D suffers with downturns... (Score:2)
There could also be R&D into shit that ultimately won't go anywhere anyways.
Re: Right, R&D suffers with downturns... (Score:1)
Most don't know anything (Score:1)
They just jump into a jupyter note book, put up some annoying graphs that may or may not indicate a correlation between some sets of data. Remove outliers and then jam it into what ever machine learning model of the month they have access to in PyTorch or Tensorflow and then get confused when it isn't magically creating good models when compared to other methods.
There are cool things that CNNs and DNNs and capable of, but often those that describe themselves as datascientists don't know how to achieve it.
Re: (Score:2)
Questionable data science (Score:3)
6% normed against what? What is the usual turnover for April and May? It's not 0 normally surely, without some context of the normal turnover, how am I to know if it's 25% higher or 100% higher, hell it might be lower for all I know.
Re: Questionable data science (Score:2)
Dead weight (Score:1)
IMHO that's probably really all it is. My wife and I work at completely different types of companies, and between the working from home and reduced hours, we've both seen how obvious it us that some people were barely doing anything at all, and in a few cases, even truly-"busy" people weren't really doing anything useful to the company. Companies have had a lot of that dead weight
Sure it will (Score:2)
My magical AI told me to buy a magic dog, a Labracadabradoodle and it worked!
Re: (Score:2)
If you rub its doo-doo, you get a free wish. However, I burned my free wish on trying to clean my friggen hand off.
"Good cultural fit" (Score:2)
'"[Companies] feel this is a time they can get rid of extra hires or lower performers who are not a good cultural fit," he adds.'
I suspect "good cultural fit" means "never challenging the boss' interpretation of things" and "never complaining about late night work calls regarding trivial matters".
Re: (Score:2)
Re: "Good cultural fit" (Score:3)
Never questioning the boss is more what Asian culture/companies do. That's why Asian companies produce some of the worst software UX that exists, basically they just design it the way the boss wants it. If the boss thinks the website needs 500 links on every page like it was 1999 all over again, then every page gets 500 links.
Re: (Score:2)
In a previous job my (academic) supervisor used to walk into my office at five pm and ask me to do a sample size calculation for something or other, rush, because the grant deadline is midnight. Typically I would produce a value, he would look at it, and say "I don't like that number. Make it smaller."
I suspect the answer to equivalent requests determines who is and is not a "good cultural fit." I am typically not a good cultural fit.
Comment removed (Score:3)
Re: (Score:1)
Reckoning vs. Rationalization, for the uninitiated (Score:2)
Sigh, yeah. Not the first time I've heard this doublespeak. Lemme take a shot at translating this for readers who might not have encountered this before.
* Reckoning: This is what happens when you finally have to face reality, due to unavoidable circumstances and evidence.
* Rationalization: This is what you do when you come up with new reasons to avoid facing reality.
Exampl
"mentoring services to data scientists" .. a what? (Score:2)
What?
Are the scientist going crazy or unable to control their urges or?
If they are "scientists" then they are (should be) the mentors.
If they need mentoring, then they are just working on data. Data annalist or processors...ok?
Am I a scientist while I just write software?
Data Scientists ~ Web Masters (Score:2)
Data scientists are just this bubble’s web masters.
You're right (Score:2)
That's not to say their job was useless or that these companies never should've hired them in the first place.
They either:
A) Never brought anything tangible to the table.
B) What they had to offer was worth the investment, now it's not.