Air Canada Has to Honor a Refund Policy Its Chatbot Made Up

https://www.wired.com/story/air-canada-chatbot-refund-policy

by gavman • 2 years ago

243 147 2 years ago

danpalmer2• 2 years ago

I'm glad to see that cases are starting to be decided about the liability of using AI generated content. This is something the general public should not need to second-guess.

bitmasher9• 2 years ago

Honestly LLMs aren’t ready for customer service. If I’m talking to a company I need to have a high degree of accuracy. LLMs are less accurate than trained humans.

danpalmer2• 2 years ago

This is my personal perception, but I think it's important that there is a clear definition of liability so that companies are able to make their own determinations of what is ready and what isn't.

denton-scratch3• 2 years ago

Me: Chatbot, are you speaking for $COMPANY? Are you $COMPANY's agent? Can I take your statements as being $COMPANY's legal position?

Chatbot: <waffle>

Me: Please put me through to a person that can articulate $COMPANY's legal position. This conversation can serve no more purpose.

dtjb2• 2 years ago

Few front-line agents have deep knowledge about their company's products or services. They trace their finger through some branches on a flowchart then dictate from a knowledgebase.

yodon• 2 years ago

Dupe https://news.ycombinator.com/item?id=39378235 (400+ comments)

ChrisArchitect• 2 years ago

Pro-tip OP: https://hn.algolia.com/?dateRange=pastWeek&page=0&prefix=fal...

quartz• 2 years ago

"Air Canada argues it cannot be held liable for information provided by one of its agents, servants, or representatives—including a chatbot," Rivers wrote. "It does not explain why it believes that is the case" or "why the webpage titled 'Bereavement travel' was inherently more trustworthy than its chatbot."

This is very reasonable-- AI or not, companies can't expect consumers to know which parts of their digital experience are accurate and which aren't.

BugsJustFindMe3• 2 years ago

Forget about digital experiences for a moment. Forget entirely about chatbots.

Air Canada argues it cannot be held liable for information provided by one of its agents, servants, or representatives

That includes EMPLOYEES. So they tried to argue that their employees can lie to your face to get you to buy a ticket under false pretense and then refuse to honor the promised terms? That's absolutely fucked.

sokoloff• 2 years ago

Fortunately, in this case, the consumer had enough proof of what happened and the court rightly told Air Canada to get fucked with that argument.

mschuster912• 2 years ago

They tried to argue that their employees can lie to your face to get you to buy a ticket under false pretense and then refuse to honor the promised terms. That's fucked.

Pretty standard behavior for big companies. Airlines and telcos are the utter worst... you have agent A on the phone on Monday, who promises X to be done by Wednesday. Thursday, you call again, get agent B, who says he doesn't see anything, not even a call log, from you, but of course he apologizes and it will be done by Friday. (Experienced customers of telcos will know that the drama will unfold that way for months... until you're fed up and involve lawyers)

FergusArgyll• 2 years ago

Standard for airlines

RajT884• 2 years ago

I have had this happen in a sense.

I once booked a flight to meet my then-fiancee in Florida on vacation. Work travel came up unexpectedly, and I booked my work travel from ORD > SFO > TPA.

Before I made that booking, I called the airline specifically to ask them if skipping the ORD > TPA leg of my personal travel was going to cause me problems. The agent confirmed, twice that it would not. This was a lie.

Buried in the booking terms is language meant to discourage gaming the system by booking travel where you skip certain legs. So if you skip a leg of your booking, the whole thing is invalidated. It's not suuuuper clear, I had to read it a few times, but I guess it kinda said that.

Anyways - my itinerary was invalidated by skipping the first flight, and I got lucky enough that someone canceled at the last minute and I could buy my own seat back on the now-full flight for 4x the original ticket price I paid (which was not refunded!).

I followed up to try and get to the bottom of it, but they were insistent they had no record of my call prior, and just fell back on "It's in the terms, and I do not know why you were told wrong information". Very painful lesson to learn.

I try and make a habit of recording phone conversations with agents now, if legal in where I'm physically located at the time.

ysavir• 2 years ago

I can sort of see it. On the one hand, it's reasonable to hold them accountable when an employee gives you the wrong discount. But if an employee, on their last day at work, decides to offer the next person calling all of the seats on a single flight for just $10, I think we'd all agree that it would be unreasonable to expect the airline to honor that offer.

It's the degree of misinformation that's relevant.

wredue• 2 years ago

This is especially true since, as it comes to refund policies, businesses make it exceedingly difficult to sort through the information.

It’s not the consumers fault that the AI hallucinated a result (as they are known to do with high frequency).

wing-_-nuts• 2 years ago

I once ordered a gift for my father for christmass. The order page indicated that it would arrive on time. When it didn't arrive, I requested a refund. They then pointed to their FAQ page where they said that orders during the holidays would incur extra processing time, and refused the refund.

I wrote back that unless they issused a refund, I would issue a charge back. You don't get to present the customer with one thing and then do otherwise because you say so on a page the customer has never read when ordering.

They eventually caved, but man, the nerve.

dataflow3• 2 years ago

This actually sounds like an interesting case to me because the details make a huge legal difference in my mind. (But IANAL, maybe I'm entirely off base here.)

E.g., did they tell you the shipping date after you placed the order, or before? If it was afterward, then it can't have invalidated the contract... you agreed to it without knowing when it would ship. If they told you before, then was it before they knew your shipping address, or after? If it was beforehand, then again, it should've been clear that they wouldn't be able to guarantee it without knowing the address. If it was after they got the address but before you placed the order, then that makes for a strong case, since it was specific to your order and what you agreed to before placing it.

bluGill4• 2 years ago

I expect employees to know the correct answers and give them to me. When an employee says something that contradicts other policy pages I'm expecting that to be a change to company policy to me - they represent the company.

If the company doesn't agree to that, then they need to show the employee was trained on company policy and was disciplined (on first offense maybe just a warning, but this needs to be a clear step on the path to firing the employee) for failing to follow it. Even then they should stand by their employee if the thing said was reasonable (refund you $million may be unreasonable, but refund purchase price is reasonable)

manicennui• 2 years ago

This line of argument is crazy and infuriating. "Air Canada essentially argued, 'the chatbot is a separate legal entity that is responsible for its own actions,' a court order said." Do they expect people to sue the chatbot? Are they also implying that people have to sue individual agents if they cause a problem?

grotorea3• 2 years ago

27. Air Canada argues it cannot be held liable for information provided by one of its agents, servants, or representatives – including a chatbot. It does not explain why it believes that is the case. In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission. While a chatbot has an interactive component, it is still just a part of Air Canada’s website. It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot. https://www.canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/202...

Real legal comedy. Since this was in small claims court maybe it was an amateur on Air Canada's side?

lenerdenator3• 2 years ago

If they could reasonably expect to be able to hire people who would agree to accept all liability incurred during their work for the company, they absolutely would.

Same with chatbots. Even better, because once it's "trained", you don't have to pay it.

There's a few instances of expecting digital entities to shoulder the entirety of legal liability here in the last few years; DAOs are another example of this in the crypto space.

JJMcJ2• 2 years ago

In early days of computerization, companies tried to dodge liability due to "computer errors". That didn't work, and I hope the "It was the AI, not us" never gets allowed either.

Aardwolf• 2 years ago

That didn't work

It worked in the British Post Office Scandal: https://en.m.wikipedia.org/wiki/British_Post_Office_scandal

hnfong• 2 years ago

Since it's a scandal, it worked until it didn't.

And AFAICT "the computer did it" wasn't the argument, it was "the computer did it so it must be correct because the experts said so".

JJMcJ2• 2 years ago

The fallout from this isn't done yet.

CoastalCoder• 2 years ago

I think this is a very different kind of thing. IIUC:

With Air Canada, the question is whether or not a chat bot can be treated as a company representative that makes binding commitments.

With the British Post Office, the issue is whether or not a software system is inscrutable during legal proceedings.

lenerdenator3• 2 years ago

They still do. Bank errors are an example.

hunter2_• 2 years ago

I don't really understand why generative LLM output is being presented to (and interpreted by) users as intelligence. This seems incredibly disingenuous, and I would prefer to see it characterized as as fiction, creative, realism, etc. -- words that make it clear to average people that while this might be entertaining and even mimicking reality, it's completely distinct from the writing of a person. Disclaimers (often small, low contrast, or "ToS;DR"-esque) are insufficient when the UI is specifically crafted to appear like chatting with a person.

im3w1l3• 2 years ago

I hope there can be some reasonable middle ground. I think in this case it's good the woman got her money. But Air Canada, presumably scared of what the next case might cost them decided to turn the chat bot off entirely. I think that's a bit unfortunate.

I don't know what the solution looks like. Maybe some combination of courts only upholding "reasonable" claims by AI. And then insurance to cover the gaps?

pjc50• 2 years ago

The "reasonable" middle ground is that chatbots should not be used unless and until their answers are reliable.

avn2109• 2 years ago

> "... turn the chat bot off entirely"

This sounds like an ideal outcome!

lgleason2• 2 years ago

If the company is not confident with the output of the chatbot or the liability then they should turn it off. All business decisions are based on assessing risk. Ultimately either insurance will need to step in or the company is willing to take the hit for mistakes their system makes because it saves them more money. This is how fraud protection it assessed with banks, credit cards etc. as well.

bgirard• 2 years ago

I agree. If bad information is given and the customer makes a decision on it, then either the customer must bear the risk or the business. I think it's more fair for the business to be out the money.

In this case this could have just been a $650 'bug bounty' had Air Canada issued a quick refund. A reasonable QA expense to find out that your AI agent is misleading your customers.

dukeyukey• 2 years ago

I don't see why chatbots can't be kept to the same standard as human staff. If an airline support agent lied to me to sell me tickets, no shit I'd want a refund and compensation! Chatbots should be allowed to be wrong, but the company should be prepared to face the consequences of that.

SketchySeaBeast2• 2 years ago

Why is that unfortunate? I honestly don't think there's any value in using AI for this if it's not guaranteed to give you correct answer.

im3w1l3• 2 years ago

I think that there can be an error rate that is low enough to still be very useful and a net-positive for the economy. But if there is a small chance to just end the company on the spot, then no one will use it despite the benefits. So how can we have the good parts without the bad, this is my question.

itopaloglu83• 2 years ago

Hmm. “Let me turn on this lead water pipeline which may or may not poison an entire town and then blame faul when it does.”

I don’t think failing to take adequate precautions is preventing AI tools from being used. I think this was plain corporate incompetence and greediness. They started using a system without properly testing it and don’t want to pay for the consequences.

What if Boeing says “Oops. We forgot to put the bolts that keeps the door in place but we shouldn’t be kept accountable for our actions”? The fact that they used a tool for it shouldn’t change the outcome unless we are going to create indemnity for big cooperations.

rideontime• 2 years ago

I struggle to comprehend how it's "unfortunate" that they turned off the customer support bot that lied to people.

im3w1l3• 2 years ago

Ideally they would fix it rather than give up on the idea entirely.

callalex• 2 years ago

Or just hire a customer service department. That worked fine for the last hundred or so years.

matthewfelgate• 2 years ago

hahahahaha

viranchee• 2 years ago

hahahhahaha

VHRanger• 2 years ago

Why are we posting this when it's just rephrasing an Ars Technica article? It's even mentionned at the bottom:

"This story originally appeared on Ars Technica."

Give the clicks to the original article:

https://arstechnica.com/tech-policy/2024/02/air-canada-must-...

hoppyhoppy2• 2 years ago

Both sites are owned by Conde Nast

emaro• 2 years ago

I don't like Ars Technica because they break reader mode and load articles chunk by chunk – I consider it hostile towards the user and am glad since Wired work much better.

chucksmash• 2 years ago

What browser? Just tried it out on Firefox for Android (version 122.1.0, with uBlock Origin enabled but JS still allowed on ars) and for the link above, I see the whole article after immediately switching to reader mode.

lgleason2• 2 years ago

Good. If you use a tool that does not give the correct answers you should be held liable for the mistake. The takeaway is, you better vet your tool. If the amount of money you loses from mistakes with the tool is less than the money you saved using it then you make money, if not, you may want to reconsider that cost saving measure.

lenerdenator3• 2 years ago

What if the company responds that they don't know how to vet the tool?

After all, we're still not 100% sure how LLMs make their decisions in what they string together as output, so the company's not _technically_ lying.

lazycog512• 2 years ago

This is why you link the source content from the RAG pipeline instead of pretending the bot knows everything.

anonu• 2 years ago

Reminds me of https://www.moralmachine.net/

floatrock• 2 years ago

Here's the real punchline:

Experts told the Vancouver Sun that Air Canada may have succeeded in avoiding liability in Moffatt's case if its chatbot had warned customers that the information that the chatbot provided may not be accurate.

Here's a glimpse into our Kafka-esque AI-powered future: every corporate lawyer is now making sure any customer service request will be gated by a chatbot containing a disclaimer like "Warning: the information you receive may be incorrect and irrelevant." Getting correct and relevant information from a human will be impossible.

unsupp0rted• 2 years ago

That's why I get all my questions answered by CC'ing legal@example.com and privacy@example.com

mathgradthrow• 2 years ago

the standard disclaimer for fraud

OnionBlender• 2 years ago

If that is such a perfect way to avoid getting sued, why don't they put that on every page of their website and train all of their customer service staff to say that when they talk to customers?

effingwewt• 2 years ago

....Because an actual judge hadn't said it yet.

I still.remember when Microsoft updated thr 360 TOS to force arbitration the day after it was deemed legal in a completely separate case.

Rest assured there is an incoming flood of TOS updates.

waveBidder• 2 years ago

Don't know if you've had the pleasure of interacting with US health insurance, but they do this for coverage and cost estimates all the time, so there is unfortunately precedent in the States.

add-sub-mul-div• 2 years ago

Humans aren't perfect but they're reliable enough to be trusted with customer service work. Humans (generally) have a desire and need to keep their jobs and therefore hold themselves to a high enough standard of performance to keep their job.

And maybe employment contracts have language that offloads liability to an employee if they go rogue and start giving away company resources. Chatbots aren't accountable in any way and we don't know yet if their creators ever will be either.

And disclaimers are used in lots of contexts too.

frud2• 2 years ago

I used to work for a company that sold factory automation technology, and had hundreds of manuals for all the products they sold. In the front matter of every manual was a disclaimer that nothing in the manual was warranted to be true. This was automation equipment running things like steel mills, factories, petroleum plants, where failures could result in many deaths and grave financial losses.

sandworm101• 2 years ago

Well, if the information on the website is too inaccurate to trust then it is also too inaccurate for purposes of contract. The logical next step is to say that all contracts made with chatbots are unenforceable as they lack any trustworthy meeting of the minds.

rnewme• 2 years ago

Because you would choose someone else for things that really matter to you, probably the same person you would chose in that kind of scenario anyway. Eg. for my initial llc where I was single employee, I didn't care about my accounting basically saying in the contract "we ll do our best but can't be liable for anything", whereas now that I have bigger llc with many employees I picked (what I perceived) the best option on the market, for a premium price. They take the liability for their work and have insurance.

rootusrootus• 2 years ago

Because it is a bad look, I assume. If I'm interacting with a company that constantly disclaims everything they say as probable bullshit, I'll go find a competitor that at least pretends to try harder.

yieldcrv• 2 years ago

or its buried on the 4th page of the disclaimer

if you read them there is often stuff like that, the most flagrant one I read said “everything above should be considered apocryphal”

deegles• 2 years ago

much like Tesla's Autopilot which cannot be responsible for an accident because you're supposed to be hands-on-wheel and alert at all times while using it.

w0m• 2 years ago

That issue is media mostly optics I think. Once you sit down with it, it's simply a better cruise control.

Now, if you are talking about 'Full Self Driving' - then yea, there's a waiver and a point there.

dheera2• 2 years ago

OR they could just honor what the chatbot said and forget the legal battle.

The chatbot's error cost them what, $200? And it probably replaced at $100000/year employee?

more_corn• 2 years ago

Also, the chatbot proposed a very reasonable solution. Book the fight, send us a death certificate when you have it and you’ll get the discount. That’s actually what the policy should be and it’s a quite reasonable error for a human to make too.

passwordoops2• 2 years ago

You think front-line support staff account for $100k in compensation?

saintfire• 2 years ago

I would assume the average salary of a customer support worker is dramatically lower than that. The point is still valid regardless.

nimbius• 2 years ago

choosing to accept an unnecessary quantifiable liability and wrapping it in a disclaimer as part of a critical business process is not a recipe for sustained growth or profit.

AC and other corporations would do well to put the brakes on this instead. identify ways to transfer risk (AI Insurance for example) or avoid risk (scrap the AI bot until the risk is lowered demonstrably.)

savvy advertisers would jump on this opportunity to show just how much AC cares about the customer and eat the loss quietly before it ever went to trial.

_thisdot• 2 years ago

Can’t they link to sources? Like Perplexity.ai or Arc Search does.

I don’t even need it to tell me anything. Links are all that is relevant. Google Analytics on the Web does something similar. You can ask questions in the search box and it takes you to a relevant page.

“Can I get refund on my flight 2 hours in advance?”

“Here is a link to refund policies w.r.t time before flight”

Sayrus• 2 years ago

Air Canada's chatbot returned sources along with its answer. Quotes from another article[1].

Air Canada’s argument was that because the chatbot response included a link to a page on the site outlining the policy correctly, Moffat should’ve known better.

[1] https://techhq.com/2024/02/air-canada-refund-for-customer-wh...

rwbhn• 2 years ago

Part of disclaimer at irs.gov for their interactive assistant: "Answers do not constitute written advice in response to a specific written request of the taxpayer within the meaning of section 6404(f) of the Internal Revenue Code."

Silhouette• 2 years ago

Presumably this is why there is a trend in consumer rights legislation towards being explicit that anything a buyer has been told by a seller or actively told the seller before the sale is material to the contract of sale regardless of the seller's small print that says "This contract of sale we wrote by ourselves and won't negotiate with you and only this contract means anything". Then they can't promise the world to get the sale and then wash their hands of any resulting commitments two seconds after they get your money. Which seems entirely fair and reasonable to me, whether the promise came from a real person or a chatbot.

AlbertCory• 2 years ago

IANAL, but AFAIK you can't disclaim liability that you actually have. I'd love to hear an actual lawyer who knows (not a know-it-all amateur) declaim on this, but:

A Ferris Wheel operator cannot make you sign a disclaimer that they're not responsible if it collapses and kills you. Or rather, they can, but it will not hold up in court.

Similarly, you can say in your manual, "We're not responsible for anything we say here" but you still are.

I don't know about chatbots, but I'd expect that judges will look for other precedents that are analogous.

astolarz• 2 years ago

Personal anecdote: A few years back I left my car at a dealership for some warranty work that was going to take a few days. It has a soft top and they left it in their gated lot overnight, where it got broken into (slashed the top, ripped open the glove box, stole a cheap machete I got from a white elephant exchange). They claimed that they weren't liable at all since I signed a waiver and should go through my own insurance. After a little push back, they caved and covered it under their insurance like they should have from the beginning. I don't go to that dealership for anything anymore, for that and other reasons.

upofadown• 2 years ago

Peter Watts comments:

* https://www.rifters.com/crawl/?p=10977

jeffbee2• 2 years ago

I don't even see how this is a big story. An expressed representation about what a product is or does, how it works, and the terms of sale in consumer contexts are binding on the seller. It is the same in America. If you go into a store and they say that you can return it if you don't like it, then you can. If you buy a TV and the guy at the store tells you it also makes pancakes, you can get your money back if it turns out that it does not make pancakes. This is true even if the representation is made by some 16-year-old kid working at Best Buy. By extension it would still be true even if it is made be an automaton.

acdha• 2 years ago

The interest in this story comes from the fact that Air Canada tried to fight it using the argument that they aren’t liable for anything their agents say. Businesses around the world are so eager to start laying people off that it’s almost certain that they will not be the only ones to try a variation on that claim. I see the important point being reinforcing liability laws and especially not allowing them to shirk responsibility with disclaimers — otherwise we’ll just see more companies cut staffing on their phone support to force people to use a system with a “you are responsible for checking everything this chat bot says for accuracy” notice.

jeffbee2• 2 years ago

I mean, they're welcome to try, and this can't be the first attempt. But there's plenty of jurisprudence at least in the USA that you can't disclaim expressed representations.

dnussbaum• 2 years ago

This type of failure is becoming more and more common as companies roll out AI systems without robust accuracy audits & human supervision.

I'm working on something to make this easier - reach out if I can be helpful (email in bio).

logicalmonster• 2 years ago

Would a company be liable to uphold its promises if a rogue human customer service agent promised something ridiculous such as 1 million dollars worth of free flights?

pizzalife• 2 years ago

No, the specific employee would most likely be liable if there is criminal conduct (this varies obviously). But a chatbot is not a person.

npongratz• 2 years ago

But a chatbot is not a person.

We live in an interesting world. In the US, a corporation is legally a person, and a chatbot is not a person[0]. I'm looking forward to the first Supreme Court case involving a corporation consisting of chatbots.

[0] I'm handwaving in this lead-in to the fantasy here, so, dear reader, please give me a break for oversimplifying and ignoring technicalities.

bluGill4• 2 years ago

The company would have to prove the human was knowingly acting outside of their job/training and was disciplined for that. Such discipline must be on the path to firing the employee if the behavior isn't corrected. Note that training is important here, an employee who isn't trained is assumed to have more authorization than someone who is.

Or in this case they need to take the AI out of service immediately until they can get a corrected version that does not do such a thing. I will accept that the AI can be tricked to do such a thing and remain in service, but only if they can show the tricks are something an honest human wouldn't attempt. (I don't know what this is, but I'll allow the idea for someone else to propose in enough detail that we can debate if a honest people would ever do that)

grotorea3• 2 years ago

Google "too good to be true contract law" and there's some info, seems the answer is "no".

thefatboy• 2 years ago

hopefully

yifanl• 2 years ago

https://decisions.civilresolutionbc.ca/crt/crtd/en/item/5254...

The resolution is n amazingly clear piece of legal writing that explains the involved thought process of the the decision and then awarding the damages. I might end up using this pattern for writing out cause and effect.

lawkwok• 2 years ago

Thanks for sharing. I really enjoyed the read and the concise decision-making and legal terms used.

spywaregorilla3• 2 years ago

The real desire here is to get it to promise a lifetime of free service.

edit: arguing that the chatbot is a separate legal entity is a wild claim. It would imply to me that air canada could sue the ai company for damages if it makes bad promises; not that air canada is excused from paying the customer.

bluGill4• 2 years ago

If I were the agent who made that promise what would they do to me? Fire me - then fire the chatbot. Put me through training - then train the chatbot. Stand behind me and do what I say - then stand behind their chatbot.

I bet they already have policies in place for this - while how they apply to AI may be different, they shouldn't let this slide.

spywaregorilla3• 2 years ago

The key term here is "apparent authority". You don't need to be granted authority for each and every thing you do, but if you start grossly doing things you were not supposed to do, that's mostly just fraud on your end.

The common example in textbooks is someone continuing to do business as an employee after having been fired. They can still make valid deals with other entities due to apparent authority if they're not clearly made out to be separated.

Cpoll• 2 years ago

promise a lifetime of free service.

That might be too wishful thinking. The tribunal would take into account damages, and whether it was reasonable to believe that you're entitled to free service.

In this case, the chatbot promised a ~$800 discount, and the tribunal awarded ~$800. But I doubt they'd make the same decision again, or deem the lifetime service enforceable/un-cancellable.

spywaregorilla3• 2 years ago

I think it's within the bounds of reason if you can cite emotional damages to grant pretty much any recourse. It maybe won't fly if you're clearly trying to trick the AI, but I think there will be general plausible deniability.

It may feel weird, but it's utterly insane to delegate customer interactions to an agent that has nobody's interest in mind, not even their own, whom you cannot trust to abide by policy.

Similarly, if the bot negotiated any sort of special deal, I think it would be very, very difficult to argue that it lacked apparent authority to make deals or that it's not a fair consideration.

jerf2• 2 years ago

This is pretty close: https://www.msn.com/en-us/autos/news/ai-blunder-chat-bot-sel...

alsetmusic• 2 years ago

My father died in hospice the night before a flight to see him. I missed the flight because there was no longer any reason to get to the airport before dawn. I called to reschedule a few hours later.

The human on the other end rescheduled and gave me a bereavement rate. She told me it was less money, but didn't mention the reason. I didn't put that together until later. She just helped me out because she had compassion.

I am too cynical to think that an AI controlled by a corporation will do this.

liendolucas• 2 years ago

Good. I hope people out there also discover chatbot holes and exploit them. Chatbots are one of the most useless and time wasters things out there, they literally serve for absolutely no purpose. And most of them work exactly like nested dropdowns where you select one option after the other. Oh and when you really want to talk to a human being in almost every scenario that option is not available. What a wonderful world powered by "AI".

stainablesteel• 2 years ago

i would expect something like this to severely stunt chatbot adoption

i think the only reason this should go through is if it didn't have a proper disclaimer at the beginning of the conversation

frud2• 2 years ago

The real story here is that Air Canada's lawyers argued, among other things, that the chatbot was a separate and independent legal entity from Air Canada and therefore Air Canada was not obligated to honor the made up policy.

In other words, this was possibly the first historical argument made in a court that AI's are sentient and not automated chattel.

martopix• 2 years ago

Or maybe it simply means that it's a service offered by another company

fnordpiglet• 2 years ago

This should hold. If AI can remind us of what humane policies are, so be it.

pier25• 2 years ago

Did Air Canada use a chatGPT for their legal defense?

Also:

Air Canada essentially argued that "the chatbot is a separate legal entity that is responsible for its own actions,"

What does this mean?

That the chatbot was provided by a third party hence they are responsible for the content provided?

Or that, literally, a chat bot can be considered a legal entity?

jerf2• 2 years ago

I think this article's full import is not being properly processed yet by a lot of people. The stock market is in an absolute AI frenzy. But this article trashes one of the current boom's biggest supposed markets. If AIs can't be put in contact with customers without exposing the company to an expected liability cost greater than the cost of a human customer representative, one of their major supposed use cases is gone, and that means the money for that use case is gone too. There's probably halo effects in a lot of other uses as well.

Now, in the medium or long term, I expect there to be AIs that will be able to do this sort of thing just fine. As I like to say I expect future AIs will not "be" LLMs but merely use LLMs as one of their component parts, and the design as a whole will in fact be able to accurately and reliably relay corporate policies as a result. But the stock market is not currently priced based on "AIs will be pretty awesome in 2029", they're priced on "AIs are going to be pretty awesome in July".

LLMs are a huge step forward, but they really aren't suitable for a lot of uses people are trying to put them to in the near term. They don't really "know" things, they're really, really good at guessing them. Now, I don't mean this in the somewhat tedious "what is knowing anyhow" sense, I mean that they really don't have any sort of "facts" in them, just really, really good language skills. I fully expect that people are working on this and the problem will be solved in some manner and we will be able to say that there is an AI design that "knows" things. For instance, see this: https://deepmind.google/discover/blog/alphageometry-an-olymp... That's in the direction of what I'm talking about; this system does not just babble things that "look" or "sound" like geometry proofs, it "knows" it is doing geometry proofs. This is not quite ready to be fed a corporate policy document, but it is in that direction. But that's got some work to be done yet.

(And again, I'm really not interested in another rehash of what "knows" really means. In this specific case I'm speaking of the vector from "a language model" and "a language model + something else like a symbolic engine" as described in that post, where I'm simply defining the latter as "knowing" more about geometry than the former.)

dlqx• 2 years ago

This mistake could have been done by a human agent as well and the consequences would have been most likely the same, wouldn't it?

zzz999• 2 years ago

Sometimes the courts get it right

Air Canada Has to Honor a Refund Policy Its Chatbot Made Up

Related Stories