When AI Doesn't Match the Hype - Are There Humans Behind those Robot Lawyers?
DoNotPay is getting more scrutiny
We are fans of innovation at Synthedia and Voicebot.ai. When fantastical claims are made about new AI applications, we would like nothing better than for them to be true. However, we also know that some entrepreneurs and many marketers will overstate their products’ capabilities in hopes of gaining perceived market differentiation, favorable media coverage, and new customers.
The “fake it until you make it” ethos is even embraced by some companies. The thinking goes that “everyone is doing it, and we will be at a disadvantage if we don’t,” or “by the time the customer buys the product, it will be true,” or “gaining customer revenue and market share will give us funding and knowledge that will enable us to make it true.” You might also hear, “I thought it could do that.”
I would rather have these companies not overstate their capabilities, but I won’t even say that it is always a losing strategy. It often is a losing strategy because a product that doesn’t work as advertised typically loses in the marketplace eventually. But, sometimes, no one checks on the false claims, the product is good enough to meet some needs, and the growing customer base does enable the product to catch up with the promises over time.
This does create issues sometimes for both companies and industries. If an industry becomes viewed as filled with scams, then even the honest and useful companies often get labeled with the same bad associations.
The Rise of the Robot Lawyers
When you log into Spellbook and ask it for an entire contact or clause, it generates it in near real-time. I’ve seen the demo of the live software. A lawyer should still review the text to make sure it is correct, but you can see for yourself that it is coherent. It also doesn’t take hours to generate.
When you request some types of legal documents from DoNotPay, you might wait several hours or a day. Latency is a thing when you consider AI model performance, but not this type of delay.
Kathryn Tewson is a paralegal at KUSK Law. She was interested in the claims made by DoNotPay and decided to try out the service that bills itself as “The World’s First Robot Lawyer.” Voicebot.ai’s Eric Schwartz commented:
“A viral Twitter thread turned Techdirt article by Kathryn Tewson provoked skepticism over whether AI had much of a role in producing complex legal documents for DoNotPay. Tewson attempted to generate a defamation demand letter and was told it would take an hour, while a divorce settlement agreement was promised in eight hours (a timeframe noted by Voicebot for the same service). That’s very slow for a generative AI system, and they still weren’t ready by then.
DoNotPay produced a small claims lawsuit filing much more quickly but included clauses unaddressed in the initial set-up and looked like a previously completed document with some of the proper nouns replaced. All three services are no longer available on DoNotPay. Tewson ended up getting a refund and was blocked on Twitter by [the company CEO Joshua] Browder.”
Tewson also found some interesting data in the documents that DoNotPay did produce. This included the name of the original document creator of an MS Word document that was converted into a PDF and the name of a recent editor. This doesn’t conclusively settle the question of how much AI is involved in DoNotPay, but it does raise questions about whether this is a service that may have some AI elements but is really driven by humans.
Voicebot’s Schwartz also evaluated DoNotPay’s new DoNotSign solution “for analyzing the long and tedious terms and conditions text that nearly everyone agrees to without reading.” He wrote:
“The ‘robot lawyer’ platform has offered the service since 2019, but the latest iteration is DoNotPay’s first to employ OpenAI’s GPT-3 generative AI model. DoNotPay CEO Joshua Browder claims the AI can spot red flags and anything unusual in the thicket of legalese. Judging from the output, GPT-3 finds legal jargon just as dull as any human would, offering unimaginative paraphrases of nearly every clause.
This is actually a good use of GPT-3’s summarization features. Whether it actually provides meaningful value is an open question, but this feature appears to be fully AI-based.
Does it Matter?
Google Duplex’s debut amazed just about everyone. Its smooth-talking humanlike assistant could contact restaurants and book you a reservation or book a hair salon for appointment. Below is the video I captured during its first public demonstration at Google’s I/O developer conference in 2018.
However, we later learned there was an army of people that were helping complete those bookings. I reported at the time:
“The New York Times reporters confirmed with Google that the call was placed by a human. ‘The company said that about 25 percent of calls placed through Duplex started with a human, and that about 15 percent of those that began with an automated system had a human intervene at some point.’ That means about 64% of all Google Duplex tasks are fully executed by the AI and 36% require some sort of human assistance.
“However, the Times reporters ad hoc testing with over a dozen restaurants resulted in four completed bookings with three made by humans. Those results suggest something closer to the inverse with 25% of calls using Google Duplex. According to the Times, ‘In other words, Duplex, which Google first showed off last year as a technological marvel using A.I., is still largely operated by humans.’”
This may also be the case for some of DoNotPay’s products. Then again, the company may have some products that use AI and some that do not. Synthedia wrote about DoNotPay’s demonstration of a bot that helped reduce a journalist’s cable TV bill. This is not a product that DoNotPay offers but the demonstration appeared to work and it definitely had AI elements.
DoNotPay started out helping motorists get out of paying parking tickets. That use case appears to have been a success. Its not clear that much AI is required for that product either. If customers are getting value, then it may not matter for DoNotPay’s business over the long-term. You would also think that humans might do a better job even if they are slower.
Marketing AI
Exclusively, marketing your use of AI is more likely to get you media coverage, customer interest, and attention from venture capital investors. DoNotPay is exceptional in generating this type of interest. “The world’s first robot lawyer” is a more interesting storyline than “a new company with online legal templates.”
I’m guessing that Kathryn Tewson’s negative sentiment is not unique, despite the many supporters of the company. The implication is that some people now believe that AI applications are not living up to the hype because of their first-hand experience. We are likely to see more examples emerge both from DoNotPay and other companies in the industry that erode trust in the technology. If ChatGPT wasn’t so popular, it would likely have a more significant negative impact on consumer perception.
It will be interesting to see whether DoNotPay can grow into its marketing messaging. The company is exceptional at securing favorable media coverage, but, for now, the robot lawyer looks more like a portfolio of online legal templates with a light smartering of AI technology and humans in the loop.
This would explain DoNotPay's sudden pivot back to customer support. Looks like the AI lawyer will not be taking the stand on February 22.