In today’s column, I examine an emerging trend involving the use of AI to serve as a stand-in at political debates. The idea is that if a candidate opts not to attend a scheduled political debate, generative AI will be employed to represent their political position. The debate then proceeds with one or more human politicians debating the AI as though it is a surrogate of the absent or no-show candidate.
This certainly seems to be a clever way to lean into the amazing fluency and conversational capabilities of modern-day generative AI. A contrasting viewpoint is that AI is not a viable or fair representation of a human candidate and thus should never be used in this manner.
What do you think about this controversial act?
Let’s talk about it.
This analysis of an innovative proposition is part of my ongoing Forbes.com column coverage on the latest in AI including identifying and explaining various impactful AI complexities (see the link here).
The Old Hat Of Skipping A Political Debate
It is well-known within political circles that not all debates are worthy of a candidate’s efforts.
There are lots of reasons to not participate in a particular debate. The debate forum might seem stacked against the candidate. The debate timing might be problematic. A myriad of factors come into play. Each situation is different and requires meritorious analysis.
To be clear, you cannot categorically say that just because a candidate opts to not agree to a debate, they are somehow right or wrong in their decision. For example, a rule of thumb in politics is that if you are ahead of an opponent then undertaking a debate with them might be an imprudent choice. The rationale for not doing so is that you are freely giving away added attention to your opponent, plus there is a risk that your standing might diminish because of the debate.
The calculus entails a kind of cost-benefit and ROI political analysis.
Does participating in the debate provide a positive net result?
If not, sidestepping the debate is probably the sensible path.
Often, a candidate who wants the debate to occur will profoundly try to shame their opponent into participating, betting that public pressure will force them to show up. If public pressure doesn’t compel them, at least there is possible political fallout for not having entered the debate. A common wink-wink recognition in the political realm is that if the circumstances were reversed, namely that the candidate demanding the debate was in the ahead role, they would undoubtedly be doing everything they could do to avoid the debate.
That’s the way the ball bounces in the wacky realm of politics.
What To Do If A Debate Opponent Won’t Show
Suppose a debate has been scheduled and one of the two candidates is all in, while the opposing candidate indicates they are not going to participate. Let’s assume for the sake of simplicity that there are just two candidates. This assumption can readily be readjusted. I’ll discuss this modification toward the end of the discussion.
What might happen if there is an anticipated no-show?
The usual result is that the debate is canceled before it takes place. The candidate who wanted the debate is bound to be greatly irked. They wanted the debate. They likely need the debate politically. Efforts to shame the opposing candidate will continue but the political gain is not usually longstanding.
Another possibility is to proceed with the debate. The podium for the opponent might be left empty. This has historically occurred from time to time and is known as empty chair debating. A famous example consisted of Clint Eastwood speaking to an empty chair.
Here’s how an empty chair debate often takes place. A moderator will ask questions of the candidate who has shown up. This is no longer considered a “debate” in the true sense of the word. It is more akin to an interview. Throughout the said-to-be debate, the candidate that is present will invariably seek to remind viewers as much as possible that their opponent is absent. Sometimes the moderator might also jump onto that bandwagon and reprimand the missing candidate.
Empty podiums don’t usually garner viewership, and the so-called debate is a bit of a dud.
Thus, a variation is to have another person stand in for the opposing candidate. This person might be an official representative of the absent candidate. That situation is relatively rare. You see, if the candidate themself is not going to participate, they seldom are willing to send a surrogate. Even if they are willing to do so, the question is whether the opposing candidate is going to take a hit in their political standing by debating someone other than their true opposition. Some would say it isn’t a proper debate if the two actual parties aren’t going directly head-to-head.
What if the opposing candidate won’t provide a substitute?
The candidate that does show up might find someone else to serve as a surrogate for the absent candidate. As you might imagine, this is highly problematic. The chosen person might intentionally try to undermine the very person they are supposedly representing. It would seem quite an eyebrow-raising circumstance to allow anyone other than the actual candidate or their designee to take the debate stage.
That pretty much covers the human-related permutations.
Generative AI Enters Into The Debate Arena
A brand-new option exists for coping with a missing candidate at a political debate. It is an option that has only become sensibly possible in recent times. Advances in technology have opened an exciting possibility or perhaps stirred a Pandora’s box.
Generative AI can be used as a stand-in at a political debate.
Boom, drop the mic.
The reason that generative AI comes to the fore is that the likes of ChatGPT, Claude, Gemini, Llama, and other major generative AI apps are incredibly adept at being able to interact in seemingly human-like ways. We all know now that generative AI can mathematically and computationally impressively mimic human writing and human speech.
Might as well use AI for political debating too.
You can log into just about any contemporary generative AI and begin a debate on nearly any topic that ordinarily comes to mind. The AI will debate you all day long. Most of the time, the AI will probably do a yeoman’s job of countering whatever political stance you might take. You can guide the AI to be easy on you or tell it to be a tough debater. If you like your debates to be mean and sour, you can instruct the AI to toss the proverbial kitchen sink at you. You decide the rules and what you want the AI to do.
Bringing Generative AI To The Debate Table
Generative AI can be deployed as a political debate surrogate in either of these two ways:
- (1) Candidate self-devised version. The candidate who intends to be a no-show opts to craft a generative AI that will sufficiently represent them at a political debate.
- (2) Opponent devised version. The opposing candidate makes a generative AI that presumably represents the to-be-absent candidate.
The first approach is the stronger of the two.
The candidate that is going to be absent opts to make a generative AI that will represent them during the debate. Presumably, the generative AI will be carefully devised to make sure that it is accurate and apt. Any blowback of what the generative AI says is going to fall squarely on the shoulders of the candidate. They will have no place to hide due to their gutsy action of putting forth the generative AI on their personal behalf.
Here’s my take. Because of a public perceived sense of strict accountability, I don’t expect many candidates to go this route. The risks tend to outplay the benefits right now. Maybe this will be viable in a few years, but it’s a frail option at this time.
Going the second route is more likely.
An opponent prepares a generative AI that represents the absent candidate. There isn’t quite the same semblance of obligation to make sure that the AI is on target to represent accurately the positions of the missing candidate. That being said, the same concerns exist as does when an opponent chooses a human surrogate without the consent of the absent candidate. Since the opponent is crafting the generative AI to represent the absent candidate, things might get slipshod or purposefully tilted. The odds are that the representation will potentially be less stellar or even possibly intentionally swayed against the actual candidate who is absent.
Significant AI ethics and AI law qualms arise in this second route, such as:
- Will this kind of AI usage mislead the public and confound public perceptions about the actual positions of the human candidate who is absent from the debate?
- Do prevailing election laws and regulations allow this or prohibit it?
- Can the absent candidate sue for misappropriation of their persona and image?
- Etc.
There is a bit of an added twist that many don’t consider when pursuing this second route. A political backlash can slam the candidate who chooses to use generative AI to represent their opponent. Here’s why. If the generative AI seems out-of-sorts, the public might criticize the deviser for playing nasty political games and being underhanded. It will seem gimmicky. It might seem overly desperate.
All in all, political players need to think mindfully about instituting generative AI in a political debate maelstrom, no matter what route is chosen.
Going Multi-Modal With Generative AI
Currently, generative AI tends to invoke a single mode of communication, often text-based.
The circumstance of a political debate would seem to involve a text-based mode whereby you enter a prompt, and the AI responds with an answer or essay (see my coverage of prompting strategies at the link here). This could work engagingly for an online-only debate. However, using such a mode during a televised debate would be tedious and unlikely to hold the attention of viewers. Imagine the boring aspects of a moderator or human candidate typing prompts into generative AI. People would tune out in droves.
The better aim would be to use generative AI that can do speech such as the nature of speaking that we are accustomed to with Alexa and Siri. Likewise, the AI ought to be set up to listen or receive audio input. A podium could be placed on the debate stage with audio equipment connected to generative AI that would be the virtual stand-in for the absent candidate.
Here’s a seemingly small detail that is likely to raise a big issue. Should the podium exhibit a cardboard cutout or a similar static photo of the absent candidate? The upside is that doing so continually reminds the viewers of who the generative AI represents. The downside is that the cutout might seem cheesy, possibly distracting.
This takes us to the realm of using full-on multi-modal generative AI and going beyond cardboard cutouts. Envision this approach. A computer screen or display is placed at the podium. An AI-crafted video as a type of deep-fake of the absent candidate is shown (for my description of how AI is used to craft deep-fakes, see the link here). The video appears to be showing the absent candidate in real-time, speaking with lips moving and their head bobbing, modeled on the same mannerisms of the actual candidate.
You can abundantly imagine that this would be quite engaging for the audience. The setup would consist of an AI-simulated animated video stream that leverages multi-modal generative AI and works in real-time. Great or dour? Once again, this might be a bridge too far when applying AI. Laws about deep-fakes might come into the matter, along with the misappropriation of the modeled person’s legal rights.
I have a mind-bending added option for you.
Are you ready?
You’ve certainly seen in the news that humanoid robots resembling human beings are being actively developed. Those could be used as a stand-in at a political debate. No need for a simpleton approach involving a mere computer screen at the podium. The robot would walk around, wave its arms, make key points via the generative AI speech capability, and otherwise appear to be acting as the candidate might.
Notably, the concerns of a robot stand-in are on par with the qualms about the other approaches, perhaps even more troubling, it would seem.
Data Training For The Generative AI
Will the AI represent sufficiently the viewpoints of the absent candidate?
That is the zillion-dollar question.
The key is the extent and thoroughness of data training used to data train the generative AI. As background, when building generative AI, the AI makers first widely scan the Internet to find various data for pattern-matching purposes. This gives them the foundational building blocks of the AI being able to converse and interact based on computational patterns of human writing.
Two major avenues of data training arise when modeling a political candidate:
- (1) Root data-training. During the initial overall scan of the Internet for foundational data training, various facets of the absent candidate might have been examined and patterned into the generative AI. This is likely by happenstance.
- (2) Supplemental data training. To bring generative AI up-to-speed on the absent candidate in an in-depth fashion, supplemental data training is undertaken. This is focused on data specifically about the candidate.
Supplemental training involves feeding numerous speeches, press releases, and other text-based content about the candidate into the generative AI. A popular method of additional data training of generative AI consists of performing RAG or retrieval-augmented generation, see my explanation at the link here.
The matter seems extraordinarily simple. Just feed as much data into the AI about the candidate as you can find. The more the merrier. Period, end of story.
Regrettably, the world is never that straightforward.
For example, imagine feeding lots of news articles as part of the data training. There are likely some news stories that deride the candidate. There might be falsehoods or misleading characterizations included in those articles. All this assorted material goes into the woodchipper of the pattern matching. The generative AI could end up portraying the candidate in a quite unflattering and mischaracterized manner.
The root content that was scanned at the initial setup of the generative AI can equally contain biased material. Will that be an appropriate way to portray the candidate? Does this spur a justifiable flap that the generative AI has been distorted to make them look bad?
The data training considerations apply to text-based modes and multi-modal generative AI too. Besides feeding text that depicts the absent candidate, you include audio materials such as speeches they have given, interviews they have undertaken, and so on. For video patterning, you would supply videos of the candidate.
Extensive testing of the generative AI must be performed, seeking to gauge how well the AI simulates or models the candidate. Due to the non-deterministic nature of generative AI, it is challenging to do fully exhaustive testing, even if wanting to do so, see my discussion at the link here.
Authenticity Is A Far Reach
Let’s acknowledge and agree on a very crucial point. No matter what you do to craft the AI, it is not going to truly be an authentic representation of the modeled person. It will be a simulation. Furthermore, there is a substantive chance that the representation will appear to be a mockery or parody. A cheap fake, as it were.
Other uncertainties exist too.
You might be vaguely aware of so-called AI hallucinations. The notion is that from time-to-time generative AI generates entirely made-up content that has no grounding in facts. I disfavor the catchphrase of AI hallucinations since it implies that AI hallucinates as humans do, a misstated anthropomorphizing of AI, see my in-depth look at the link here.
In the setting of a political debate, suppose the generative AI suddenly and unexpectedly makes up something that the candidate never said, and which is totally unlike the candidate. If this happens in real-time, the guffaw will undoubtedly undermine the use of the generative AI. It is the bad apple in the barrel. No matter how much the rest of the AI seems to be responding suitably and representing the candidate, the glitch is going to undercut anything else the AI says as the candidate.
A complicated quagmire is also being established, as per these weighty questions:
- Do you have the AI mimic the personality and actual speaking tones of the candidate or be more plainspoken?
- What if the AI hallucinates and generates an off-the-wall answer that resembles what the candidate might have said and is inadvertently assumed to be something they once did actually say?
- When the AI says something presumably bearing the opinion of the candidate would we want to have a justification or explanation of how the AI came up with it?
- If the topics during the debate go beyond the data training of the AI, is it fair to have the AI compose a reply that might have little if any basis regarding what the candidate has previously said?
- Would people try to quote the AI as though the candidate really said what was generated by the AI?
- Would people falsely believe that the AI was a fully authentic recreation of the candidate?
- Should the AI continually remind that it is AI and not the candidate?
- Does a heated debate make the other party look odd that they are debating an AI system rather than a human?
- Should the AI be programmed to respond heatedly?
- Should the AI be instructed to be emotional, angry, indignant, happy, sad, etc.?
- Would it be best to allow each side a separate turn, or have them be able to talk over each other during the debate?
- And so on.
We don’t yet have across-the-board accepted standards on what the rules will be during a human versus AI stand-in debating forum.
The Future Is Coming Soon
I’ll wrap things up for you and provide some final thoughts for now.
One often-used argument is that if a candidate won’t show up for a particular debate, they are by default creating a vacuum and it is perfectly fine for an opponent to craft a generative AI stand-in. That’s the risk the person takes for being absent. They get what they get.
That kind of argument is probably not going to hold water in the court of public opinion.
The public is likely going to perceive the use of generative AI as a clever ploy, but the novelty is going to be overshadowed by whether the AI is suitably fair and balanced. Those who favor the absent candidate are going to cry foul. Those who disfavor the candidate will to some degree relish the use of the AI if the AI weakens the standing of the candidate. Imagine though that if the AI does a bang-up job, those same people will be mad as hornets.
Interestingly, a real-world example is soon going to play out. In Virginia, a congressional race includes a candidate who has created a generative AI to represent his opponent on a debate stage. The opponent has indicated they aren’t going to participate in the debate. Will the devised AI chatbot be a winner-winner chicken dinner or will its use create a fracas?
No worries, I’ll be covering the debate and the outcome, so be on the look for that upcoming coverage.
A closing remark.
Envision that an entity calls for a political debate and none of the candidates want to attend. There could be two, three, four, or any number of candidates. The entity on its own devises generative AI simulations for each of the candidates. A televised debate takes place with no human candidates present. It is exclusively AI-based candidates battling on the debate stage.
Mull over that astonishing situation.
Is that a political milieu we want to have occur?
Well, whether you want it or not, you can expect this is going to happen. There will be efforts like this, which might skirt AI ethics and AI law limitations and seem outrageous or unfair.
Just always remember, as they say, all is fair in love, war, and politics.
