Can Artificial Intelligence Replace Workers?

I read an article a few weeks ago about how Microsoft is cutting dozens of news workers and replacing them with artificial intelligence (article). Microsoft reported that the work of the reporters has been semi-automated for months.

This is nothing new. Many news outlets have been relying on AI to generate their content. The process generally involves an algorithm going through lots of data (likely other news stories) and laying the information out in article form. The AI tends to write in a very formulaic structure and style, mostly providing the readers with names, dates, times, etc. (Wikipedia).

It’s not clear immediately whether this is a good trend or not. Some argue that AI frees journalists from having to write mundane articles about basic current events, but some journalists find it demoralizing that they can be replaced by a robot.

According to an article by Wharton, many large news outlets are using AI in some form, including The Washington Post, The Associated Press, BBC, Reuters, Bloomberg, The New York Times, and The Wall Street Journal (article). In 2018, China’s Xinhua News Agency went so far as to use artificial intelligence to power its news anchor. The anchor was created using computer graphics and fed information via algorithm. I can’t confirm that the anchor is still working, but they also debuted a female version in 2019. There are videos on YouTube here and here.

I think the loss of jobs in this sector is a natural progression in the transfer of basic knowledge work from people to machines. The AI can write faster, cover more topics, and contribute more efficiently to our 24-hour news bombardment.

Bombardment? That might have sounded harsh, but it’s true that our news outlets have been generating far more news over the recent years. We have access to news constantly, via television news networks, websites, and through places like Twitter. How can we possibly keep up with the demands for news? It shouldn’t surprise anyone that we would need to turn to some form of automation to satisfy our needs.

The big question that we should be asking is: who’s next? Knowledge workers are potentially under fire from automated sources. I have been saying for years that people responsible for transactional knowledge work are the next logical targets.

Accountants, for example, could be replaced by AI. Many currently use options like TurboTax to satisfy their basic tax needs, when will that expand to include more complicated matters, like audits? Financial planners are also potential targets. The average family doesn’t need complicated financial advice. Make a budget, start saving, get life insurance, these are basic bits of advice that can easily be replaced by a robot. Will people feel comfortable dealing with a machine? Maybe not, but how long until we create a machine that can pass as human? Probably not long, especially since lots of this work can be done over the phone.

I’m beginning to rant a bit here, but the point is that these types of changes are not merely coming, they are here. We need to start thinking about what other jobs will be replaced and what hurdles will need to be crossed for the replacement to happen.

For example, for AI to write news it needs to understand how humans read and it needs to understand what words we expect to see. It’s not editorializing yet; it doesn’t need a lot of empathy. It just needs to find the facts and lay them out in a way that we expect.

But to replace an accountant or a financial adviser (or a lawyer), AI will need to make people feel comfortable talking about money (and ultimately fears and aspirations around money) with a computer. To do that, AI will need to be more empathetic and learn how to interact with people on a more human level.

Who knows how long it will take to get to that point?

Exaggerated Capabilities of Artificial Intelligence

I recently read an article about Gemma Milne in Forbes (article). She wrote a book called Smoke & Mirrors. It’s a book about the misuse of technical terminology and how it can affect funding, policy-making, voting, and other things.

Though I have not read her book (and, therefore, cannot vouch for it), I thought it was an interesting point that I have come across in my own business. I believe the term ‘artificial intelligence’ is way overused.

I get it, AI is cool. It makes us think of far-flung sci-fi action. A world of self-driving cars and human-like robots. That’s why it’s attached to all sorts of products that may technically qualify as AI but are far from the technology we are picturing.

I think this is due to the misunderstanding of what artificial intelligence is. AI is an over-arching term that encompasses all forms of machine learning software. And a lot of that software is underwhelming when compared to the sci-fi fantasy we picture in our minds.

I’ll give a direct example, I used Casetext software on a trial basis. They promote their legal assistance software, known as CARA A.I., as being research-oriented artificial intelligence. Literally, it’s described as “like having a research assistant at counsel’s table.” I was told that I could upload a brief into CARA, and she (or it, I suppose) would find all of the relevant cases and predict arguments that could be made. I was skeptical to say the least.

The fact is that Casetext was actually a pretty good tool. It worked like I would expect a research tool would, but it was a far cry from a research assistant at counsel’s table. I tried CARA and was, frankly, underwhelmed. I uploaded a brief. It managed to find most of the citations but not all of them, and it made no real effort to predict arguments myself or opponents could make.

That’s not to say it was bad. It wasn’t. It worked fine. I just think it was massively oversold. And that’s my point.

Artificial intelligence just isn’t at the level being promised, certainly not in a form available to consumers. Most of these tools are some form of machine learning algorithm designed to aid in automation. Why aren’t businesses more honest about the capabilities?

Because ‘machine learning algorithm’ is not sexy.

I only tried Casetext because of the AI capabilities. Had they been honest, I wouldn’t have used the trial at all. So I suppose their ruse worked, kind of. It certainly backfired when I was underwhelmed by the actual product.

The problem is that AI has been plagued by overpromises since it’s inception. For example, the term ‘artificial intelligence’ was first used at a conference at Dartmouth College in 1956 hosted by Marvin Minsky. Minsky was one of the original thinkers in the AI movement. In 1970 Minsky said that we would have artificial general intelligence (meaning humanlike intelligence) within three to eight years.

Fifty years later, we still don’t have artificial general intelligence, and there is no real estimate for when we will have it.

I’m sure you’ve heard about how lawyers will (or have been) replaced by AI, and accountants, engineers, and doctors are next. Essentially every knowledge worker is on the chopping block.

But the technology is nowhere near capable of doing basic jobs like driving trucks, let alone taking on complex knowledge work. Don’t get me wrong, our day will come. There is reason to believe that AI will be able to do these jobs in the future, but it’s much harder than these companies will let on.

With AI’s capabilities still so limited, companies should stress the advancements we have made. We have seen great leaps in natural language processing for example, which helps streamline search engines and research software (like Casetext), and games have been dominated by AI over the last year (consider AlphaGo and AlphaStar for examples). That’s really cool and should be celebrated.

Let’s move away from the hype.

Paycheck Protection Program Flexibility Act

The Paycheck Protection Program (PPP) has come under some scrutiny over the last few months. Many business owners felt that the program didn’t give them enough flexibility to use the funds productively while maintaining their right to have the loans forgiven.

The government listened and the Paycheck Protection Program Flexibility Act (PPPFA) was signed into law on June 5, 2020. It contains some much-needed changes to the original program.

Only 60% of the funds need to be used for payroll

The original PPP loans required businesses to use 75% of the funds for payroll. Most businesses found that to be way too restrictive. The issue was that PPP assumed that most businesses had payroll costs that were about four times their rent and other costs. However, that is not true for businesses in expensive areas. And many businesses had laid off their workers due to the closing. PPP required that they be rehired before June 30, 2020, but in the meantime, many businesses had very low payroll costs. Therefore, businesses were concerned that they wouldn’t hit that 75% requirement.

Business advocate groups were asking for 50% to be spent on payroll (article), but reducing the required percentage to 60% should give most businesses the breathing room they need. Frankly, the funds were meant to protect paychecks (hence the name), not keep businesses open, so requiring that a portion be spent on payroll makes sense.

The funds must be spent in 24 weeks

The PPP required all funds be spent in eight weeks. Businesses complained that they would prefer to use the funds when the economy reopened rather than use them on idle workers. This extension gives everyone until the end of 2020 to use the funds, which is hopefully enough time for the economy to reopen.

Though, even if the economy does not reopen entirely by the end of 24 weeks, these loans were not huge so most businesses should have no trouble using the funds on approved sources (like payroll) given that timeframe.

Businesses must rehire employee by December 31, 2020

Originally, businesses that took PPP funds had to rehire any lost full-time staff by June 30, 2020. I had some issues with this requirement because it ignored whether there was any demand for products or services. Meaning, a business would have to staff back up without any demand. Why would that make sense? My concern was that businesses would just shed those employees by the end of July. There would be no reason to keep them.

Now businesses have till December 31, 2020, to rehire employees. That seems like plenty of time to me. It also addresses the other problem: workers don’t want to return. This is partly due to the fact that workers on unemployment are making more money than if they were working. That additional unemployment incentive is fading away in August, but I’ve heard talk that there may be an additional payment of $450/week being offered to help phase people back to work.

The time frame allowed by PPPFA should be long enough to get past these incentives to stay home.

There have also been changes to the requirements to rehire the same number of full-time employees. Under PPP, a business had to return to the number of full-time employees that it had as of February 15, 2020. There were essentially no exceptions, other than the employees didn’t have to be the same employees you had previously.

Now PPPFA states that if you can’t get back to your previous number of employees, your loan can still be forgiven if you can show one of three things:

  1. You were unable to rehire an employee that worked for you before February 15, 2020;
  2. You are able to demonstrate an inability to hire a similarly qualified employee(s); or
  3. You are able to demonstrate an inability to return to the same level of business activity as before February 15, 2020.

It’s not clear how one would demonstrate that any of the above requirements have been met. The first could be shown pretty easily, but the other two will be challenging for some businesses. A restaurant will have little trouble showing that they haven’t been open, but what if they’ve been open for outdoor service? That could get tricky.

The loans are extended to a five-year term

The original loans were to be repaid over two years. The term has now been extended to five years. The first payment is also deferred until six months after a determination of forgiveness is made.

To be honest, I don’t know if they are even considering forgiveness applications, yet. The first payments will likely not be due for a long time.

PPP was only designed to cover 4-8 weeks

The main issue with PPP was that it was designed to be a stop-gap measure. Politicians at the time did not know how long this pandemic would stretch. I remember the good ol’ days when we debated whether the closures would last four weeks or six weeks. At more than double that time, the closures have taken a much bigger toll on small businesses. PPP just didn’t (and couldn’t) foresee the depth of this problem. It’s great to see some changes are being made to ease the strain on small businesses.

One other thought, businesses who got funding in the first round of the PPP loans have probably spent their money, or at least most of it. How does this benefit them? Well, the rehiring rules are relaxed, for sure, but these changes would have been better had we seen them earlier.