ChatGPT Speak-First Incident Stirs Worries That Generative AI Is Getting Too Big For Its Britches

Spooky times are here.

It isn’t even Halloween yet and already something has happened via generative AI that has gotten people a bit alarmed. The widely popular ChatGPT began starting conversations with users, including asking questions on topics that were personalized to the person being hailed.

Puzzled on why this is newsworthy?

The reason this seems hair-raising is that most generative AI is devised to wait for the human to initiate a conversation. When you log into generative AI, there is customarily a blank prompt window that allows you to get interaction underway. The screen is waiting for you. If you don’t type something, nothing happens. A conversation starter of one kind or another resides squarely on your shoulders.

Think of it this way.

If you’ve ever used Alexa or Siri, you realize that it is up to you to engage those natural language processing systems. For example, you might say “Hey, Siri” to get the AI going. This puts humans in control of things. You feel empowered when you summon the AI which then does your bidding.

Turns out that the AI maker of ChatGPT, OpenAI, has reportedly acknowledged that the speak-first issue briefly existed. Speculation ensues that this is a perhaps being worked-on feature that will soon be adopted across the board. Others claim it was just a bug. All in all, the matter could be labeled as a no-harm, no-foul situation.

Well, except for those users who thought either a ghost was reaching out to them via generative AI or that maybe AI had finally reached sentience. Some were likely scared out of their wits. Others might have been thrilled to no end and proclaimed that they knew this day would someday arrive.

Answering Questions On The Eeriness

Due to my ongoing coverage of the latest AI trends via my Forbes column, see the link here, I was engulfed with questions about what this AI speak-first phenomenon foretells.

Let’s take a quick look at a few of the most frequently asked questions and my answers.

• Does AI doing a speak-first mean we are on the cusp of artificial general intelligence or AGI?

No, it does not.

The deal is this. You frankly cannot make any interpretation either way about whether speaking first entails a sign of AGI arising. Here’s why. It is very easy with conventional everyday AI to make the AI become the initiator of conversations. All you need to do is establish a system prompt in generative AI that tells the system to open conversations by stating something to the user. Voila, speaking first is underway by the AI.

That being said, if we reach AGI, the odds are that such AI will at times be the dialogue initiator. We will undoubtedly become accustomed to this. Right now, whether you opt to start an interaction, or the AI does, there really is no particular bearing regarding the attainment of AGI.

• Why would today’s generative AI know about me if it does do a speak-first?

Apparently, these were users who had previously made use of ChatGPT.

The significance is that when you use generative AI and carry-on conversations, the prompts you’ve entered and what you have discussed are usually recorded within the AI system. Few users realize this is occurring. If you take a close look at the licensing agreement of the AI system, there is almost always a clause that says the AI vendor can retain whatever has been entered.

Furthermore, and this is truly a scary aspect, the AI maker often reserves the right to inspect your prompts by hand, along with reusing the data for additional training of their generative AI, see my discussion about privacy intrusions concerning the use of generative AI at the link here.

I would venture odds that the generative AI tapped into prior conversations with the impacted users and merely acted as though the conversation was still underway.

• Why don’t we already have AI that does a speak-first act?

Actually, we do have this.

There are generative AI apps that are built to do a speak-first act. One of the most popular uses for speak-first involves chatbots that provide mental health guidance, see my coverage at the link here.

You might opt to have an AI app on your smartphone that is intended to advise you about mental health. Since you could be busy at times and not actively run the app, it is set up to automatically periodically ping you about how you are doing. This might happen once per day or on some other basis.

The gist is that the AI will initiate a conversation with you. The simplest starter would be “How are you doing?” while a more tailored opener might be “You’ve been having some anxiety recently and maybe it would be helpful to chat about how you are doing.”

Conclusion

Please go ahead and put down the pitchforks if the news about generative AI speaking first got your dander up.

AI makers have been cautious in permitting or devising their generative AI to be overly friendly or seem to be in your face. They realize that users might react negatively and abandon using such AI. That would be a crushing blow for an AI company, and they are unlikely to make such a brash move that ultimately shoots their own foot.

Here’s a twist for you. With so many generative AI apps to choose from nowadays, perhaps AI makers that instigate a speak-first enactment would stand out. Be on the watch that this might become a trend.

If that does come to pass, I’d vote that this gets underway for Halloween. What a great time to introduce the “shocking” approach to the public at large. Not that everyone will be positive about it. Best to be ready for a panic reaction by some that their AI has become eerily possessed.

Well, then again, maybe it is.