Artificial Intelligence – My Mental Model

Business idea: The Anti-Telemarketer

Not sure how the telemarketing scene is evolving elsewhere but in my city, I’m receiving about 2-3 telemarketing calls a day. More than half are scams. It’s getting annoying.

So I thought I’d come up with a solution.

Not long ago, Google debuted its Duplex AI. This AI assistant is capable of navigating phone calls. It’s pretty darn amazing. If you’ve yet to see it, here’s a link:

So it looks like Google’s new AI assistant is going to be capable of a whole host of tasks, perhaps even handling telemarketers. Here’s what I’m envisioning:

When your phone rings and you see a number you don’t recognize, feel free to let your AI assistant answer the call for you. If it turns out to be a legitimate call and someone you’d like to speak with, your AI should be able to navigate that conversation quickly and effectively. Perhaps your phone would ring a second time with the AI alerting you with who the caller is and what they’re looking for. If, however, it wasn’t someone you wanted to talk to, the AI assistant would know how to navigate those calls as well.

For all the calls that you don’t want to answer, let’s put them in 3 categories. First would be legitimate calls that you just don’t want to answer for whatever reason. Maybe you’re hanging out on a beach with your friends over the weekend and your boss is calling… probably about those TPS reports. Your AI assistant can let your boss know that you’re currently unavailable, take a message, and say that you’ll return that message at your earliest convenience. Or maybe it’s a call that you’d like to take but you’re in the middle of something important. Whatever the case may be, the AI should be able to navigate these conversations well enough to pass along a message.

The second category would be legitimate telemarketers. We’re talking about legitimate businesses reaching out for cold sales or surveys. For these, perhaps your AI assistant would know which businesses you have accounts with to better understand which promotions you might actually be interested in. Rather than you having to go through the whole phone call to find out what the actual pitch is, your assistant could navigate that conversation and turn it into a brief text message for you. For everything else, the assistant could quickly and politely say that you’re not interested, and request that you be taken off their call list.

The third category, and the one that inspired this idea, is the scam-based telemarketers. Fuck those guys. This week alone I’ve received:

Calls from places like Burundi, Somolia, Samoa, Seychelles, and Kalamazoo. As I understand it, the calls hang-up before you can answer. When you call back, you’re charged for your time on that call.
Calls from China telling me that I’m in big trouble relating to real estate purchases and government corruption.
Calls telling me that I’ve been busted for tax evasion and I need to reach out to my local tax office immediately.

Each is a robo-call, meaning that their process of generating leads is full automated and requires very few resources on their part. My approach has been to pick up the calls and to hang up as soon as I recognize what it is. But it’s not much of a solution as the calls keep arriving. They might even be increasing in volume. So how exactly do you fight back?

This idea is inspired by a TED Talk I saw a while back. The speaker was being solicited by an email scammer. Something to the effect of the Nigerian princess scam but it had to do with gold bullion, if I remember correctly. The speaker, like most of us, was able to pick up on the scam rather quickly. But rather than ignore, he thought he’d engage with the person on the other end for some fun. The email chain became rather entertaining as the speaker was able to get the scammer to use some questionable ‘code words’ in their communications. At the end of the day, the speaker did this because he knew that for every minute this scammer spent targeting him, was a minute he wouldn’t be able to spend targeting someone else. I appreciate his efforts as I’ve attempted the same thing… but there has to be a better way. Enter Google Duplex.

Imagine that tax-scam robo-call running into your AI assistant:

Robo-call: “The reason behind this call is to notify you that we have registered a criminal case against your name concerning a tax evasion and tax fraud in the federal court house. So if you want any further information about this case, please press 1. If we don’t receive a call from your side, please be prepared to face the legal consequences, as the issue of tax is extremely serious and time-sensitive. So have a blessed time. ”

AI: “1”

Scammer: “(In a thick Indian accent) Hi my name is Nicky Johnson, how may I help you today?”

AI: “Hi, I received a call today about owing some taxes and I’d like to pay them before I get in trouble”

Scammer: “Thank you for calling. This is a very urgent matter and we need to resolve it quickly before you’re forced to pay any penalties.”

AI: “Thank you so much for letting me know. What do we do next?”

Scammer: “Can I start with your name and social security number?”

AI: “which name?”

Scammer: “Your first and last name, and your social security number”

AI: “OK. Sure, but which one?”

Scammer: “What is your first name?”

AI: “Hue”

Scammer: “And your last name”

AI: “Jazz”

Scammer: “Thank you Mr. Hue Jazz. Now may I have your social security number?”

AI: “Which one?”

Scammer: “Your social security number, sir. It’s 9 digits and on all your tax filings”

AI: “Oh, OK”

Scammer: “Do you have it sir?”

AI: “Have what?”

Scammer: “Your social security number”

AI: “I don’t know what that is”

Scammer: “Never mind sir, we can proceed without it.”

AI: “What?”

Scammer: “We need to receive payment as soon as possible to avoid putting a lien on your assets. I can walk you through that now.”

AI: “OK”

Scammer: “To make a payment, you’ll have to go to our website. Do you have a computer in front of you?”

AI: “Yes”

Scammer: “The website is http://www.-”

AI: “Is that an upper-case WWW or a lower case www?”

Scammer: “It doesn’t matter, both will work”

AI: “OK. Do I need internet for this to work? I don’t think this computer has internet.”

Scammer: “Yes you will need internet. Do you have a computer that has internet?”

AI: “Yes, but I’ll have to start it up. You don’t mind waiting do you?”

Scammer: “No, that is OK.”

AI: Thanks, it’ll just take a few minutes”

(few minutes passes)

Scammer: “Is your computer ready”

AI: “Not yet.”

(few more minutes passes)

Scammer: “Ready now?”

AI: “Almost.”

(few more minutes passes)

Scammer: “Sir, your computer should be ready by now. Are you sure it’s working?”

AI: “Not sure. The screen is still black. Can you help me fix my computer?

I’m not a very vengeful person, but something about this just makes me all warm and fuzzy inside. The reality is that you could customize your AI to give a variety of answers and drag the conversation out in all sorts of entertaining directions. The best part though, is that for every minute they’re engaging with your AI, they’re not scamming someone else. If enough people took this approach, it would shift the balance of power. I can’t imagine people would be all too keen to put the resources into a scam-based call center knowing that 99% of their leads were AI assistants just messing with them.

While it would be fun to call this the ultimate solution to telemarketing and over-the-phone scams, it’s not. This would be a brilliant solution to the current mode of telemarketing and over-the-phone scams but the situation would evolve. The most natural evolution I can think of is the scammers and telemarketers switching over to a Google Duplex-esque robo-call. But if it’s your Google Duplex AI Assistant that is answering the calls… it becomes AI VS. AI. I’d take a front row seat to that…

The Future of Advertising

Part of my degree was in marketing so I tend to see advertising through a different lens than most. The other part of my degree was in psychology so I can’t help but see the psychological component as well. One dynamic which I find particularly interesting is in how we actively tune out the noise for traditional advertising, but when we see someone we respect engaging positively with a brand, we take note. The Starbucks cup or the Lululemon tote bag may be a bit played out today, but both are great examples of how this works.

Years ago, I was watching TV and noticed how the characters on the show had to avoid mentioning specific brands – likely because they didn’t have approval and seeking approval would’ve required legal paperwork and perhaps a fee. In certain instances, it actually made for rather awkward speech. In reality, we actually reference brands and products on a regular basis, in our regular speech, for the sake of accuracy. Which means that not being able to use brand names in certain areas of media actually hurts the dialogue. Wouldn’t it make sense to write the dialogue as you would naturally, and then approach the brands mentioned in a positive manner for ad revenue?

There are certainly some complexities to this strategy, but I doubt they’re beyond our ability to solve. Based on how I’ve seen things progress, I think this is actually being done now to some extent. What a concept, letting a character talk about their Mercedes or BMW, let them talk about their iPhone or Android, let them talk about their favorite restaurant or coffee shop. Script writers would have to maintain integrity so that it didn’t come off as a plug or mini-infomercial, but I don’t think that would be too difficult. The idea isn’t about sneaking an advertisement into something we’re already paying attention to, it’s about letting a brand impression exist where a brand impression would already naturally exist.

So maybe we’re starting to turn that corner, but where does it go from here? I have an idea.

Right now, AI and computer vision allow YouTube to recognize most copy-written material and then defers action to the original owner. As AI and computer division develop further, they won’t just be able to recognize the content, they’ll be able to recognize what’s in the content. Watching a movie, and see a sweet car that you’ve never seen before? No problem, hit pause, hover your mouse over the car and see a few quick details. Super interested? Click on the details and you’ll head straight to the website. Now imagine being able to do that with clothes, foods, toys, and everything else.

If we approach this correctly, I can’t help but think it would be a massive win-win for everyone. No more advertisements. No more commercials. No more jingles. And especially no more manipulation of public perception in the hopes of earning a sale from someone who doesn’t actually need or want your product. If this done correctly though, I think the biggest winners may actually be the businesses.

Rather than guessing at where to advertise, how to advertise, and how much to spend on advertising, just paying per click. Every time someone sees a piece of media that includes your product and someone wants to know more about it, there’s your point of monetization. Next-level pay-per-click advertising. Effectively, you’re only paying to connect people to your product, when they’ve shown an interest in your product. Not only is that a more streamlined approach, it builds trust rather than degrades it.

Efficiency is my North Star. When someone sees something they’re interested in, they want to know more about it. If they learn more about it and they want to buy it, they want a quick and easy way to complete that transaction. Businesses want to provide those details and the option of that transaction to potential customers, however, they would prefer to only spend their advertising budget on people who are interested. This strikes me as a remarkably efficient approach compared to what’s out there now.

Thinking it through a little further, I know there are bound to be a few hiccups. What happens if someone you don’t like is wearing or using your product? What if you’re just getting started and you need to get your product out there to begin with? I could come up with a few other issues that would exist in today’s unspoken rules of advertising but I can’t help but think that it’s just not that complicated. If you’re on the alt-left and someone on the alt-right is wearing one of your products? Grow up. Appreciate the extra revenue, and appreciate that if they’re wearing your stuff, you may have more in common with them than you might think. Just getting started? Send free products out to influencers who would appreciate them. If you have solid product, they’ll show it off and you should end up with a cascading effect. If you send your product out to the people who would appreciate it, and they don’t? Maybe you picked the wrong influencers, or perhaps your product just isn’t very good. Regardless of what obstacles I come up with, the solutions don’t seem very far away.

I’d estimate we’ll have the tools to do this within about 10 years. Whether or not major industry players are interested in challenging the status quo is a different story though. But this is why ‘revolutionary’ has become the holy grail of doing business. Whoever breaks that mold, I’m rooting for you.

A Completely Automated Business

Here’s a thought…

how far are we away from kids at Harvard coming up with a fully automated business for a class project?

I said it in that context for a couple:

I suspect fully automated businesses already exist in the fringe but I’m talking about something more widely applicable
By the time the kids at Harvard are doing it, it’ll be big news and government regulators will have to start shifting around this potential

Imagine a company called Widgets.com. Widgets.com is a market place for widgets where manufacturers of widgets can list their products. Buyers of Widgets can come to the site, pick the widget they want and place an order for delivery. When a customer places an order, the order is relayed through to the manufacturer and the manufacturer will ship the widget directly through to the buyer.

Widgets.com outsources it’s live chat and call center. And their web design. And their IT. And their Legal. And their Accounting. And their digital marketing.

Widgets.com would also have extensive data analytics that would help track key information for making strategic decisions. These data points would include customer feedback and reviews, website activity, error tracking, legal reporting, financial reporting, and social media stats. And anything else you wanted to include.

The Widgets.com algorithm would be capable of making executive decisions, but would aim to outsource nuanced details. For example, if pink widgets were trending on social media, a note would go out to the digital marketing team and manufacturers of pink widgets, while a request would go to the web designer to feature pink widgets. If the situation was more nuanced, say with a zero star review, the algorithm would track that info, forward it to a capable customer service rep and have them work to resolve the issue.

It’s actually a fun exercise because you can do this with just about any decision being made within a company. I’m pretty sure these are the steps to building this decision engine:

Identify the cues to look for when identifying a problem
Use additional cues to verify the problem
Review past solutions to the problem or similar problems
If a past solution has worked, use it again
1. If a past solution works again, make a note
2. If a past solution doesn’t work, go back to step 2
If a past solution didn’t work, look to variations of solutions to similar problems.

I know that’s a bit of an oversimplification but what I’m getting at is that with enough time and insight, a top CEO could effectively upload his decision engine into a neural net. Perhaps a decision engine wouldn’t make the best CEO for a complex company that operates in a rapidly changing environment with an actively engaged customer base… but maybe at that point, a human CEO isn’t cutting it either.

That’s where I see this going, especially because it’s already happening. Big data analytics is an early stage version of human/digital hybrid CEO. Right now, we’re mostly using data analytics to provide the human CEO with more information. If the human CEO sees that everything X happens, Y needs to happen, he can automate it. Once it’s automated, that’s the responsibility of the digital CEO. As more information starts to get tracked, more patterns will emerge, and more automation will occur. As that process progresses further and further, the human side is needed less and less.

I’m not sure how this will play out, but I do know that today’s pundits are suggesting that the CEO’s role will be among the last to be taken out by automation. That before the CEO role goes digital, manufacturing will be replaced by 3D printing, warehouse workers will be replaced by robots, delivery drivers will be replaced by automated trucks and drones, and even computer programmers will be replaced by computers who programmers have taught to program. Considering that the new Atlas looks like its about to try out for Cirque, who knows.

Non Biological Intelligence

Had a thought.

It was that artificial intelligence was a misnomer because biological intelligence is no more real than non-biological intelligence. I would be wrong.

While my statement would be correct in reference to the concept of intelligence, it was incorrect in reference to the world artificial I grew up understanding artificial to mean fake, not real, or an imitated version of. Turns out it’s a little more nuanced. Artificial simply means created by humans instead of occurring naturally. I’m happy I looked that up.

I had another thought.

At this stage, AI is being largely designed and furthered by humans. That workload is starting to shift. At a certain point, AIs will be able to further their own intelligence and will require no more from humans than humans require from computers today. At this point, if you can no longer say that an AI’s intelligence was created by a human, is it still artificial?

The definition of artificial suggests two things: created by humans and not occurring naturally. Once an AI takes over the development of its own intelligence, is that development not happening naturally? When naturally is defined as without special help or intervention, the answer is yes. But what about when we consider the definition of nature? Well, Google would suggest that nature is the phenomena of the physical world […] as opposed to humans or human creations. While you could make the argument that a self-developing AI was originally created by a human, it would like someone having planted a seed saying that they created a tree. And just like the tree, the next generation of offspring would lack any direct connection to human creation.

So by definition, artificial intelligence becomes non-biological intelligence once it becomes responsible for its own intellectual development. Very interesting.