Category: Software

The Ethical AI

~ Veselin ~ 1 Comment

I find it funny that every time a very clear statement is explained or exaggerated with an adverb, the explanation is a hint that the opposite is present or the statement is not entirely true.

He absolutely doesn’t drink alcohol. I will totally buy a ticket for the next Taylor Swift concert. The students will use ChatGPT entirely for research purposes.

ChatGPT and “(Ethically)!” in one sentence, in a paid ad.

The first time I encountered the moral dilemma around the use of AI was in the Robots series by Isaac Asimov. I read that long before I owned a computer and totally bought the idea of a positronic brain. Asimov saw that robots if allowed to do whatever they wanted would just start killing. He envisioned a set of forced limitations that AI never hurts humans (the full list of 3+1 laws is here) as the only way for robots to be useful. Asimov noticed in his books that robots would replace human labor and eventually cause stagnation but that was only partially addressed in his series after centuries of expansion.

Who could’ve imagined that the first appearance of any resemblance of AI would need a very different set of laws than Asimov’s first 3? The present-day AI already appears in multiple forms, each of which has its own ethical challenges. The prompt-based tools tend to use human content and present it as their own with no citation or link to the author. They’re awesome for faking homework. The image generation tools copy artists’ work and make it semi-unique, filling the need for cheap illustrations on spam websites that would slip undetected by Googlebot. The chatbots and robocalls automate tasks that were once reserved for humans, causing unemployment. There has to be a fine line between what’s okay and what isn’t.

I’ve been thinking about how I would regulate that thing since my last post on the subject from last year and came up with roughly this:

  • Any statement by AI should cite the sources of information and provide links
  • AI should not present slight modifications of human content as its own
  • AI should not use the prompts of one human outside of the context of the interaction with that human.

But after writing this, I had a lightbulb moment. If anyone put a thought on this, it has to be the EU administration. And yes, the EU agreed on a much longer document, where Generative AI is just one type of risk and contains an Asimov-like masterpiece:

Generative AI, like ChatGPT, would have to comply with transparency requirements:

  • Disclosing that the content was generated by AI
  • Designing the model to prevent it from generating illegal content
  • Publishing summaries of copyrighted data used for training
EU AI Act

Generative AI is not even considered a high-risk type of tool. EU considers AI tools as high-risk if they classify humans, analyze emotions, collect faces, provide advice on health or legal matters, talk to children, and so on. How didn’t Asimov think about that? The existential dangers of a toy that can explain dangerous activities to children.

Overall, the definition of an ethical use of AI is taking some shape but I wonder how much damage will be caused to human content creation and creativity until any of that’s adopted.

At least, none of these risks is Skynet and Asimov’s laws are not yet relevant.

Promotional pricing

~ Veselin ~ Leave a comment

I just finished the first part and here we are, a promotion for the continuation is waiting for me.

The MarTech person is me is really curious about the conversion rates of the negative discount. Does the higher price compensate for the drop in conversions by people who get annoyed by the bug?

Thoughts on Content Aggregation

~ Veselin ~ 1 Comment

TL;DR – ChatGPT is the last occurrence of a long trend for content aggregation and using other people’s content without linking the source. Reasoning:

Aggregated content is worth money

The modern human generates content all the time. It’s very clear that we do it when we post photos on Facebook or write lengthy text on our personal blogs but it’s not just that. Our browser history is content, our search history is content, and our cookies are worth something to the advertisers. We generate content by clicking the TV remote, and most likely by even speaking in the presence of a smart device with a microphone. We generate content every time we click on our phones beyond the password screen. This content is aggregated and transformed to be used by whoever can convert it to revenue with some privacy-related exceptions.

Once an engine has a database of aggregated content, it can attempt to monetize it by finding consumers looking for parts of that content, or ideally by monopolizing an audience in a specific area. Big data, stored in a way that allows quick access is like a black hole, curls the space around it and makes things happen that would otherwise be impossible. This doesn’t change this content’s nature – it’s aggregated from external sources and can’t exist without these sources. Google, for example, produces very little public content.

For the majority of the existence of the Web, the market for such aggregation was dominated by tools that would also link to the information source and share the traffic or the profit so the information source survives. Wikipedia demands a source for everything and the sources are part of each article. Google links to websites. Foursquare would link to that nice restaurant’s website. Some services would directly share revenue. We grew up with this approach and it sounds fair. But it seems like it is going to be challenged again.

Aggregators are becoming the source of truth

I observed some questionable developments around aggregation over the last 5-10 years. Google, for example, has been motivated to keep clicks within the service and it shows. They built an information source called Knol, developed a mechanism for hosting the entire web called AMP, built Google Maps, and integrated summaries in the search results. I can now learn everything about my favorite actors, for example, and see their photos without ever leaving Google. I can make 10-15-20 content-related clicks and still never hit one that leads outside of Google.

When individuals do that very same thing, it raises eyebrows. People would copy/paste and do slight modifications for homework, write a paper, trick crawlers, farm Karma on Reddit, or who knows what else. Other people have tried and succeeded in authoring books with slightly modified content from other sources. Rewording text, translating it from a foreign language, and editing photos can make them hard to trace back to the author has been a practice that’s frowned upon and sometimes challenged with legal actions.

Now that ChatGPT appeared, “AI” is the new big thing. It does not look like an intelligent bot to me, though. It looks like the ultimate copy/paste engine, no wonder it’s so good at writing homework. It has no own knowledge but it appears like it knows everything. It successfully uses other people’s creative efforts and then shares it like it just knows it out of nowhere, not citing the sources or sometimes citing without providing links. It has the knowledge, just chooses not to share, unless asked. I’ve not checked how many people work on it, I would assume thousands, but I doubt any of them are content creators. Expert scrapers – probably, experts in aggregation – likely, big data – most certainly, experts in human language processing – absolutely.

Consequences of this trend

In case ChatGPT completely replaces Google, the traffic to the original creators will decrease, although thanks to Google’s tactics, it might not decrease by much. Why should a visitor read a lengthy blog post if a bot can present a brief summary of that effort without even mentioning who created it? The aggregators and the consumer are both benefitting, it’s just that the content creator is now turned into a free “trainer” of someone’s bot. We’ll all start consuming AI-rewritten text until that breaks too. Given that AI’s text has no creativity, will, or its own ideas, every time one of us consumes it, it diverts thought and effort away from the actual creation effort.

The trend is concerning but I doubt we can do much about it. Nothing is more powerful than an idea whose time has come. ChatGPT is part of the negative trend of aggregators claiming ownership of people’s content but it may also be convenient. It’s convenient in the same way as The Pirate Bay – it has most movies created with that small issue that the service disregards the will of the people who created the movies. Just like The Pirate Bay, Google, Bing, and ChatGPT can’t exist without the people who created the underlying content they use to generate all these clicks.

I personally hope that people will push back against AI’s content rewrites and focus on services that are fair. I also hope that the disruption that’s coming will crack the near-monopoly over search. Some good things might grow in the cracks, or may not.

Poor Man’s Bitcoin

~ Veselin ~ Leave a comment

The communism withdrew from Bulgaria in 1989 and when the political police became unemployed, all sort of weird new things popped up to fill the gap. Grocery stores and supermarkets. New TV channels. More than one kind of ice cream. Fortune tellers. Horoscopes. Insurance racket. Chainmail too – you rewrite this letter 5 times and put it in 5 mailboxes and you’ll live a long and happy life. You don’t and you’ll die in pain. Multiple testimonials included.

One kind of chain mail had a price tag and wasn’t supposed to work, but worked for a while. Let’s call it The Poor Man’s Bitcoin. Here is roughly how it worked, excuse my faint memories for any inaccuracies.

There’s a sheet of paper, cut from a notebook, handwritten by a person who we can call “The Seller”. That sheet of paper contains the rules of the chainmail and 6 home addresses or PO boxes. Rules are as follows:

  • You need to send 2 Leva to the last 5 people in the chain, and also 2 Leva to the person who invented it (the number varied)
  • The way you prepare new copies is by rewriting the sheet and filling your name at the bottom of the list of 5 people in the chain, removing the first one
  • It contains terrible curses that will reach you if you violate the rules, sell more or less than 5 sheets of paper, or don’t mail money to the people in the chain
  • In order to get your money back, you need to temporarily become “The Seller”.

If you follow the rules closely, you should quickly get your money back – just find 5 people to buy your sheet of paper and letters with money would start flowing. Does it sound familiar?

  • It inflated algorithmically
  • Its intrinsic value was zero
  • The only things that kept it living for some time were people’s feelings and beliefs, mostly greed and fear of missing out
  • Some people loved it and were willing to fight that they’ll become rich once they get back their thousands of letters
  • It worked because people didn’t really understand how it works and look past the first 5 or 25 people who would give them money
  • Each transaction benefitted the author of the scheme + the regular system that guaranteed money transfer (the post)
  • Everyone could start their own chain letter

Once most of us were exposed to it, the number of letters turned out to be disappointing, and most people realized that these curses that guarantee the distribution don’t work, it vanished to be replaced years later by lottery tickets. Same feelings but no need to be able to write.

I find it amusing that so many people consider Bitcoin a form of investment while it’s pretty much like the chainmail from the 90s. At least this is how I see things. I do not understand how it works.

If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.

Active Listening

~ Veselin ~ 3 Comments

A colleague is talking. Their case is important and the more they say, the more I’m eager to add to their point or present arguments against. At some point, they need to take a breath and I feel like it’s my turn with all these things I wanted to say. I shoot and feel like something is not right.

Listening is Hard

In our culture, the opposite of talking is more like waiting to interrupt.

— Verbal Judo: The Gentle Art of Persuasion

There are many problems with the waiting to interrupt approach, for example:

  • People are not great at expressing themselves and often don’t say what they mean so replying to what they say is pointless because they don’t mean it.
  • Our brain stack is limited, we can barely remember last 2-3 things we wanted to say and if we don’t get the chance, we might feel frustration or other negative emotions. This is especially true if it’s a group discussion and 2-3 people are more vocal than others and better in interrupting.
  • We are biased when interrupting and don’t give equal chance to everyone to speak.

“Recent research has indicated that the average individual listens for only seventeen seconds before interrupting and interjecting his own ideas.”

— Chapman, Gary D. The Five Love Languages: The Secret to Love that Lasts.

We can do better than that.

Verify What We Heard

The main takeaway from Verbal Judo is that we need to listen with the goal to understand the meaning first, not with the goal to say these 3 or 5 things that came to our minds. We need to verify the meaning and an easy way to do that is by paraphrasing and asking for confirmation or clarification. “So you say (paraphrase here)?”. Note that according to other sources, this will not lead to the best outcome if we expect a tip 🙂 In that case paraphrasing should be more like a direct repetition of what we heard.

My ex-boss 8 years ago loved to stop by my desk and start asking for immediate changes to the e-commerce website we were developing. He usually had a point for these changes but it was super critical to understand what result he wanted, not what change he wanted. What he said might be “move this text from here to here”, while what he meant would be “users don’t see this cool new feature, we need a way to promote it”. Knowing the service better I was often able to suggest other approaches to achieve that same result that worked better if I managed to decode the intent of the request.

Hangouts and Meetings

Radical Candor has an entire chapter dedicated to how we can survive hangouts with the best possible outcome, which would be that all ideas are heard, challenged, and discussed, and decisions are made. The main takeaway for me is:

Give the quiet ones a voice.

— Kim Scott, Radical Candor

Without that, we’ll only listen to those 3-4 loudest people over and over, not that they’re not good. But others might be better. I’ve experienced a situation in which the two loudest people reached a quick agreement for a decision for which other people had important information but had not expressed it. Challenging a decision instead of an idea can be taken personally and hard words started flying. That’s a failure in communication that could’ve been avoided early by listening.

If all goes well, we listen, we verify and clarify the meaning of what is said. It’s our turn now to talk.

Our Turn

Chris Voss thinks that we should push for our beliefs and I somewhat agree.

Remember, pushing hard for what you believe is not selfish. It is not bullying. It is not just helping you. Your amygdala, the part of the brain that processes fear, will try to convince you to give up, to flee, because the other guy is right, or you’re being cruel.

— Never Split the Difference: Negotiating As If Your Life Depended On It.

I’ve seen so many times how a single person with a voice and a good idea can change, well, almost anything.

Earthsea

Some of the best thoughts on listening I’ve read come from Fantasy. The author is Ursula Le Guin and the quotes are said by her characters Ged and Ogion.

For a word to be spoken, there must be silence. Before, and after.

To hear, one must be silent.

Thanks for reading!