In one recent posts in this series, I argued that the latest forms of AI will play a unique role in the history of humanity and technology. Un this one, I want to drill down on that idea by showing how we’ll increasingly treat generative AI as coworkers, collaborators and more.
AI as Flawed Tool
One of the ironies of today’s generative AIs like ChatGPT is that, in many ways, they make for lousy tools in the traditional sense. What you expect from a good tool is consistency, dependability, durability and accuracy. At least for now, the today’s generative AI, especially the large language models, often fail to meet these criteria.
As I said in my last post, “If we held these AIs to the same standards as the literal tools in our toolboxes, we’d probably toss them. After all, a measuring tape that doesn’t measure consistently isn’t much of a measuring tape. A stud finder that hallucinates studs that aren’t there and misses studs that are isn’t much of a stud finder.”
Let’s get into some of the problems.
Top Five HR Functions
If you ask the generative AIs the same question multiple times, they may well give you a different answers in different instances. For example, let’s say I ask one of these AIs, “What are the five most important HR functions?”
I asked Bard this question three times. It gave me the same answer the first two times and a different answer the next day. ChatGPT gave me the most consistent responses, while Bing performed more like Bard: giving me two virtual identical answers and later a somewhat different answer.
Generally speaking, though, the five most common answers were:
- Recruitment and Selection
- Training and Development
- Performance Management
- Compensation and Benefits
- Employee Relations
This is, of course, a subjective question, so who really cares if Bard throws in “outsourcing” and Bing throws in “culture” or “talent management” sometimes? Well, not me, unless I’m trying to create a training module that needs to teach a consistent lesson. I’m not saying that issue can’t be fixed, even with generative AI, but the point is that these AIs have an unpredictability that must be taken into consideration by users and developers.
The Forces of Nature
In contrast, these AIs are much better at consistently providing information that has been well codified, such as scientific information. For example, they will consistently say that there are four forces or nature and identify them correctly. The definitions may be slightly different from response to response, but generally speaking they’ll be the same.
Undependable and Inaccurate
I have experienced AI “confabulations” many times. I’ve seen these AIs make up names of fictional scientists, tell me stories about things that could not have happened, and just get the facts wrong about basic things such as chronological order.
In my last post, I gave a detailed account of AI hallucinations and inaccuracies in regard to the topic of a famous poet. I’ve also experienced AI getting basic mathematics wrong. In fact, as I was write this, I asked ChatGPT to multiply two four-digit numbers. Not only did it give me the wrong answer twice, it gave me two different answers to the same problem!
This is common for these AIs, so when I hear that ChatGPT will soon be responsible for things like bookkeeping, I have to shake my head. The firm that carelessly turns its finances over to generative AI has best be prepared for a visit from the IRS.
That Could Change
Of course, what’s true today may not be true tomorrow. ChatGPT may become flawless at mathematics as its maker, OpenAI, forges alliances with firms such as Wolfram|Alpha. By using plug-ins and APIs, ChatGPT might be able to go from mathematical moron to savant.
Still, my point remains. Without a lot of testing, do not assume the responses coming from one of these AIs are accurate. And, if you’re purchasing an external system, be sure the vendor of the software that utilizes generative AI has a very sound explanation of how the system will be made consistently accurate and dependable.
AI as Intern
So, if these AIs are still pretty shaky as tools, what good are they? Well, that depends. What do you actually want from them?
Let’s say what you really want right now is someone who can give you a hand in a professional field in which you have pretty good skill sets yourself. Basically, you want a bright but green intern who is eager and works with remarkable speed.
In my case, for example, I know a few things about how to write a good survey about HR and work-related issues. So, I want a first draft of a survey on the topic of employment law compliance. Here’s an example of the kind of prompt I could use:
Please write a 10 question survey on the topic of employment law compliance. I’m going to survey HR professionals on this topic.
Then let’s say it gave me 10 questions that are pretty good but, because I’ll need to run a multiple regression analysis, I want some Likert-scale questions. So this time I write:
Now please provide me with five more related questions that use five-point Likert scales
And voilà! I now have a 15-question survey I can use as good draft on which to build. Knowing my job, I can spot problems and make improvements. I am, in the end, grateful for my splendid intern.
AI as Collaborator
Now let’s say I have less use for an intern than for a bright coworker with whom I can collaborate And let’s say I’ve been asked by my Chief HR Officer to look into the issue of pay transparency, which no one in on HR our team has a good handle on yet. So, I turn to a virtual coworker who, while obviously not an expert in employment law compliance, is quite knowledgeable about a wide range of issues.
Collaborating with Bing
So, I ask this virtual coworker (let’s call it Bing in this case) some questions starting with “What can you tell me about pay transparency laws in the US?”
Bing responds as follows:
“Pay transparency laws are regulations that require employers to be more transparent with salary ranges and benefits, and they aim to help promote fairness and equity in the workplace³. Depending on the jurisdiction, employers may be prohibited from asking candidates for their salary history, or they may be required to disclose a wage range for current and/or open positions². Some states and local jurisdictions also have pay data reporting requirements that mandate employers to report pay data by gender, race, ethnicity, and other categories².
“According to the U.S. Department of Labor, pay transparency can help improve pay equity, trust, talent acquisition, and employee engagement¹. However, there is no comprehensive federal pay transparency law in the United States, and the laws at the state and city level vary widely in their scope and enforcement³⁴⁵. You can find more information on federal and state-level equal pay and pay transparency protections for workers on this map¹.”
Then Bing provides me with the following sources that I can use to verify what its saying:
(1) US Pay Transparency Laws by State 2023 | PayAnalytics. https://www.payanalytics.com/newsroom/us-pay-transparency-laws-by-state.
(2) Pay Transparency: What It Is and Laws by State – ADP. https://www.adp.com/spark/articles/2023/03/pay-transparency-what-it-is-and-laws-by-state.aspx.
(3) Equal Pay and Pay Transparency | U.S. Department of Labor. https://www.dol.gov/agencies/wb/equal-pay-protections.
(4) The State of Pay Transparency Laws: What You Need to Know … – LinkedIn. https://www.linkedin.com/pulse/state-pay-transparency-laws-what-you-need-know-2023-aspenhr.
(5) Pay Transparency Laws By State  – Zippia. https://www.zippia.com/advice/pay-transparency-laws-by-state/.
Batting Ideas Around with Bing
Once I’ve studied the sources, I learn that, although there is no comprehensive federal law in this area, it might make more sense to create a national policy since our company has major operations in California, New York and Illinois, where there are existing or pending laws.
So, Bing and I bat this idea around, listing the pros and cons of a national policy. Ultimately, with Bing’s help, I put together an internal report on the existing laws and the advantages and disadvantages of a national policy. Ultimately, although I’m collaborating with Bing, I’m responsible for the report and must verify every fact Bing provides.
From here, I know this is going to become an organizational decision that involves lots of different stakeholders as well as legal counsel. But my collaborator and I have done a pretty good job, I’d say, of quickly putting together a balanced report that will help the company make the best decisions about a complex issue.
AI as Contractor
Now let’s say I’m part of a small communication team that’s tasked with communicating our company’s guidelines and best practices around hybrid and remote work. The policies are already developed so our job is to make them as accessible, educational and engaging as possible.
One way to do this is via visual elements. Our team brainstorms some possible fun illustrations but there’s not much that matches our vision in the clipart files and there’s no budget to contract with an outside artist on this project.
A member of the team says she might be able to use one of the new AIs to generate the illustrations we have in mind. By the end of the day, she’s shared 40 different images with the team, and we select 6 for the guidelines document.
Someone makes the comment that he wished all their graphic artist contractors worked so quickly and cheaply. This gets a bit of nervous laughter. After all, as writers, we’re well aware that the large language models work a lot cheaper and faster than we do.
AI as Dance Partner
Ultimately, these generative AIs don’t easily fit any pre-existing categories. Technically, they are tools but historically unique ones. Because of this, it often makes more metaphorical sense to view them as playing roles more similar to other human beings, with known strengths and weaknesses.
There’s the role of smart and fast intern who, nonetheless, is prone to making potentially serious mistakes. There’s the role of a eager collaborator who brings many talents and total open-mindedness to the table. You can bat ideas around with this person but, ultimately, you will be responsible for the outcomes of that collaboration. And, of course, there’s the role of contractor with special skill sets.
In all cases, though, there needs to be a growing familiarity with these AIs as they become regular “dance partners” in the workplace. You must get to know their tendencies and cadences, and you are responsible for taking the lead in whichever virtual dance you’re doing. Because, although these tools will certainly be used for automation, they are best at augmenting and complementing people with existing skill sets.
Or, at least, that’s how things stand today. Who knows what tomorrow brings?