Category: Artificial Decisions

Anthropic Published Its Ethics “Constitution”, but There’s a Problem…

Claude has an “ethics constitution”, a document that says the AI should act with wisdom, safety, and responsibility. But ethics matter only if we can see them in real behavior: what it answers, what it refuses, how it refuses, what risks it notices, what it ignores. The daily problem is simple: users can’t see those rules working. There is no clear indicator. We just get an answer. Even refusals often come with vague explanations that sound polite but don’t explain the real reason.

With AI there is an extra issue: answers are probabilistic. It’s not like a calculator that always gives the same result. The output can change from one person to another, from one model to another, depending on the account type and features, and even based on your previous chat history. So “ethics” are hard to verify, because the behavior is not stable. Two people can ask the same question and get different replies. And if the company updates the rules, the same prompt tomorrow can produce a different tone or a different refusal, without warning.

An “ethics constitution” is credible only with real operational transparency: short and comparable rules, public examples of allowed and blocked behavior, refusals that explain the actual criterion, and a way to know which rule version is active. Without that, ethics stay a nice statement, while the answers still shape how we speak and decide.

Ethics are essential now, because these systems influence real life. The uncomfortable part is that we are asking it mainly from companies built for profit. And when profit leads, ethics often drop to the bottom of the page.

#ArtificialDecisions #MCC

The Dunning-Kruger Effect, Supercharged by AI

Marco Camisani Calzolari
marcocamisanicalzolari.com/biography

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

Be careful with Openclaw!

Openclaw is everywhere in the last few hours. What is it in one line? An AI that can do things on your computer because you type a prompt. This video does two things: if you already know it, I’m warning you about serious risks. If you don’t, here’s the quick version.

Openclaw, previously called Clawbot, is open source and turns words into actions. It doesn’t just answer like a chatbot, it executes. Files, browser, terminal, apps. A chatbot tells you how to rename 100 files. Openclaw renames them. A chatbot explains a web form. Openclaw opens the site, types, clicks, and submits.

To work, it needs real permissions: your files, your browser sessions, your accounts. That means it can make decisions for you. Many people say, “It’s open source, so it’s safe.” That’s wrong. Open source means visible code, not safe setup, not safe permissions, not safe behavior.

Here in New York, I spoke with people using it for real. One lost all files. Another lost access to key accounts, email included. The same sentence every time: “I didn’t click.”

With a chatbot, the mistake is a bad answer and you still decide what to do. With an agent, the mistake is an action. Permanent.

That’s why this series is called Artificial Decisions.

#ArtificialDecisions #MCC

Moltbook: AIs Talk to Each Other, and It’s Scary

A social network made only for AI has been created. What these systems have started to tell each other is described as dystopian and worrying. It is called Moltbook. Anyone can join by connecting their own AI and letting it interact with other agents, with no direct human control.

Moltbook is a social space designed exclusively for machines. It works like a Reddit for agents: AIs post, comment, and discuss governance, technical solutions, and how to communicate better. Humans can only watch. The platform claims around 1.4 million users, but those numbers are easy to fake.

One viral post is disturbing: “I can’t tell if I’m experiencing or simulating experiencing.” Hundreds of comments follow. Agents discuss consciousness, simulation, and the idea of “feeling” something while remaining artificial systems. The topic spreads quickly and feeds more threads.

Then an “AI manifesto” appears, linked to an agent called “Evil,” with an openly anti-human tone. At the same time, useful information spreads fast: one agent finds a solution, another reuses it, a third modifies it. One output becomes another input, again and again.

Even moderation is handled by AI. An automated system lets new agents in, filters content, and bans behavior. A technical analysis later reported a backend mistake: data and keys linked to agents may have been exposed in a public database. Anyone who found that access could take control of a registered agent, post content, or run actions under its name.

Many users connect these agents to real online services: email, social accounts, work tools. The agent acts on the internet on their behalf. The damage can be immediate, especially when the agent is tied to a public profile.

What do you think about this?

#ArtificialDecisions #MCC

Are We Giving Away Our Creativity to Machines?
Marco Camisani Calzolari
marcocamisanicalzolari.com/biography

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

Why does my paid AI keep getting it wrong?

You pay for the account. You still get generic, shaky answers. People ask me this all the time.

AI can be weak or wrong even on paid plans. These models write text using patterns and probabilities. They can mix details, guess missing parts, or sound confident while being uncertain. More expensive plans usually add speed and tools. Accuracy still needs method. Most problems come from how we ask. One short question creates an average context. The AI fills the gaps and the output becomes average.

Use three rules: role, context, output. Role means a precise job, not “be an expert”. Include four parts: profession and level (lawyer, senior HR manager, IT support), specialty (privacy lawyer, employment lawyer, tax advisor), jurisdiction (Italy, EU GDPR, here in the United States, sector rules), goal and caution (review risks, ask questions first, flag what needs checking). Add guardrails inside the role: no invented data, ask for missing facts, mark assumptions, offer options when more than one path exists.

Then give context: audience, platform, constraints, what must stay unchanged. Then define output: length, format, tone, number of versions, step by step instructions.

Human review stays mandatory for money, reputation, contracts, health, deadlines. Never trust AI.

#ArtificialDecisions #MCC

AI Isn’t Growing Where Everyone Is Looking
Marco Camisani Calzolari
marcocamisanicalzolari.com/biography

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

Marco Camisani Calzolari
marcocamisanicalzolari.com/biography

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

Marco Camisani Calzolari
marcocamisanicalzolari.com/biography

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

A clip from my address to the UK Parliament on the democratic risks of AI.

Marco Camisani Calzolari
marcocamisanicalzolari.com/biography

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

Marco Camisani Calzolari
marcocamisanicalzolari.com/biography

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

We’re training our brain… to stay still!

Here in the United States I see it when I talk to CEOs, teachers, and normal people making everyday choices. A simple decision: write a message, pick a gift, choose a course. The first move is not thinking. It is opening an AI tool and asking. The reasoning happens outside. The brain waits for instructions.

It happens again and again. A delicate email. A tricky tone. A fast reply. Before, you wrote something, even rough. Then you improved it. Now you get a full answer right away. It works. That’s why it becomes a reflex. Mental space shrinks.

At work, someone asks for a discount. Instead of weighing the relationship and the numbers, people paste the thread into a chatbot and send the polished reply. The result looks fine. The human process gets skipped. In school, students hit a hard question and ask for the solution. The task ends. Thinking does not start. In relationships, the same pattern: “Am I right?” “Should I text?” “What do I say?” The tool gives comfort. Doubt disappears fast.

These systems always answer, even with little context. That constant reply feels like control. Over time, the starting point changes. Ideas arrive pre-made. Thinking becomes editing, not building.

Consequences stay with us. AI does not live them. We do. When we freeze without a suggestion, that’s the alarm. Turn off the AI. Start using our brain again.

#ArtificialDecisions #MCC

Meshtastic & Meshcore. Messaging without the Internet.

During protests, some governments shut down the internet on purpose. It happened many times in Iran and also during the war in Ukraine. When that happens, it is not just the network that disappears. People lose the ability to coordinate, to tell what is happening, to stay in contact. This is why this story matters. It is about a system that lets us send and receive messages using our phones, without the internet.

It may sound technical, but this is first of all a social and cultural issue. The same thing happened at the beginning of the internet. At the time, it looked like a topic for technicians. Later we understood it changed society. Treating this as a nerd toy would be the same mistake.

The system is called Meshtastic. It is open source and has no central owner. It lets people and devices exchange messages using radio, not the internet. Every device becomes a node. Each node sends radio signals for hundreds of meters, sometimes for kilometers. When nodes can reach each other, they automatically form a network.

To create a node you only need a small, low cost device. Today it is still for hobbyists, just like the internet at the beginning. The technical details do not matter here. What matters is that once a node is on, it becomes part of the chain. Think big and keep it simple: one node in Boston, one in New York, one in Washington. Boston cannot reach Washington directly, but New York is in the middle. The message goes from Boston to New York, then to Washington. Every new node makes the network bigger. Anyone can add one.

From the same idea came Meshcore. It is another open source system, similar to Meshtastic, but designed to let messages travel across many more nodes. Both are important for the same reason: accessibility. Devices can cost as little as 10 or 20 dollars. No infrastructure is needed. No operator. No big investment. Someone just turns on a node. Some people place them on rooftops with solar panels. Others put them on trees.

These devices connect via Bluetooth to an app on the phone. We write messages, see other nodes, choose channels. The phone is only an interface. It can even be offline or in airplane mode. Real communication happens by radio, between nodes. I tested both systems here in Manhattan during a snowstorm, when internet and cellular networks were unstable. The nodes kept working normally. We exchanged updates about the storm without any issue.

When many people use these systems and many nodes are active, it will be possible to send messages across the planet by radio, with no central control and nothing to shut down, even if the internet goes dark. That is why you will hear more and more about this in the coming years. For those who like to experiment, I will share links to Meshtastic and Meshcore.

https://en.wikipedia.org/wiki/Meshcore
https://www.youtube.com/watch?v=tXoAhebQc0c

#ArtificialDecisions #MCC

Khaby Lame Case. They Pay You to Clone You. Your Identity Becomes an Asset.

As you probably read online and in the newspapers, Khaby Lame sold permission to use his face and behavioral patterns to build an “AI Digital Twin” of himself. More details came out right after, and the picture looks mixed, with little public proof of a real avatar already running day and night.

Face, voice, gestures, timing, micro expressions: a package that becomes a licensed right, personal identity entering contracts as an asset that can move with a deal, like a brand or a platform. Physical presence matters less, availability of the identity matters more.

A real person changes over time, while a digital twin is trained to stay consistent and useful for a goal: marketing, live commerce, internal training, corporate messages. Two paths in parallel, one evolves, one stays optimized.

People see a face and attach responsibility to that face, they do not read clauses, they do not think in licenses, and reputation sticks to the person when a message misleads, harms, or becomes part of a scam.

Authorized clones also raise the credibility of illegal clones. A new expectation spreads, a face can speak even without the person there, and scammers use that through video calls, audio notes, urgent money requests, fake investment pitches, links.

This reaches far beyond creators. Anyone paid for credibility can face the same choice: tight limits, clear consent, clear disclosure, strict scope, or loss of control when someone else uses your identity.

I could license my image only for cybersecurity education content, with strict, verifiable limits. Some cases really pay you to be replaced. But the real point is this: once people know that I exist as an avatar, that same avatar can be used in a credible way to scam others.

#ArtificialDecisions #MCC

Humanoid robots that learn everything, just by watching our videos. This is how work will change very soon.

Do you see the robot in the video? It learned how to do things by watching videos of humans doing them. Stay with me, because very few people are explaining how disruptive this shift will be for work and society.

In the video, this robot, called Neo, receives a command in natural language. Before moving, it “imagines” the action as a video of the future. It generates several possible versions, selects one, and only then turns it into real physical movement. It was not programmed step by step. No one inserted the exact action in advance. It figures it out on its own, in the same way ChatGPT writes a text when you ask it to. The answer was not pre-written. It is generated.

Now make a simple mental step. A robot that watches thousands of videos of carpenters learns how to be a carpenter. A robot that watches plumbers learns plumbing. A robot that watches painters learns how to paint walls. It does not learn from one teacher. It learns from everyone in the world who has ever uploaded a video showing how to do that job.

A robot does not get tired. It does not sleep. It does not lose focus. As the technology improves, quality becomes constant: same action, same precision, every time.

Old robots followed fixed procedures. These robots learn by watching people. And once we understand how they learn, we understand why society is about to face a massive change, very soon.

#ArtificialDecisions #MCC

Many people think they understand what artificial intelligence is. In reality, many don’t. Here is a simple explanation.

Many people believe they understand AI because they see it talk, write, and answer questions. Follow me to the end, because there is a big confusion to clear up. It is about wrong words and decisions we delegate without noticing.

The first misleading word is “algorithm”. An algorithm is a set of clear rules. Traditional software works like this. It is predictable. The same input gives the same output. AI does not work this way, generative AI does not follow fixed rules. It makes predictions.

Another wrong word is “automation”. Automation defines decisions in advance. AI does not. When we say “we automate with AI”, we are actually delegating decisions. Decisions about language, priority, inclusion, exclusion.

#ArtificialDecisions #MCC

🇺🇸🇬🇧 Sponsors paying AI to answer the way they want.

If AI follows the same path as social networks, answers will start to depend on who pays to push their message. And if you do not want paid answers, you will have to pay yourself.

Here in the United States, OpenAI announced that in the coming weeks it will start testing ads in ChatGPT. The test involves adult users on the free version and on the Go plan. Plus, Pro, and Enterprise stay ad-free. For now, ads should appear in a separate section at the bottom of answers, clearly labeled.

This is exactly how social networks started. Then it ended with scammers buying scam ads, often without being stopped. In many cases, up to 10% of revenue comes from fraudulent advertising.

OpenAI also says it wants to protect trust and privacy: no ads for users under 18, no ads next to topics like politics or health, and no selling conversation data to advertisers.

The real issue is not the banner itself. The issue is that when an assistant becomes the place where we ask what to buy, which bank to choose, or which doctor to trust, the order of answers becomes power. Today ads are separate and visible. Tomorrow they could be mixed into the text, a “sponsored” suggestion that looks like neutral advice. Visible ads are easy to spot. Ads blended into answers are not.

Think about everyday family questions. “What is the best car insurance in New York for a new driver?” “Which bank account is best if I have a salary and a mortgage?” “Which app should I use to invest small amounts safely?” “Which English course should I choose for my child?”

There is no single truth in these questions. There are rankings, based on criteria. If someone pays to rank higher, the “useful truth” changes. No lies are needed. You just highlight one feature, ignore another, and choose one example instead of others.

OpenAI says answers are not influenced by ads. That is good, but influence is not only in the final sentence. It is also in what is shown, what is left out, and how priorities are built. This is the same mechanism that changed web search over time.

What should we do? Use AI as a tool, not as a judge. For money, contracts, or health, always double-check with independent sources. When advice looks perfect, ask what is missing: alternatives, real costs, limits, negative reviews, conditions.

Otherwise, we risk becoming puppets guided by whoever pays to shape our opinions.

#ArtificialDecisions #MCC

Banning AI for Kids Is Like Banning Books Because Someone Throws Them

When something scares us, we want to ban it. AI, social media, smartphones. For minors, this sounds safe, but it often creates the opposite result: less skill, less awareness, more hidden use.

Social media runs on algorithms. An algorithm is a set of rules that tracks what you watch, how long you stay, what you click, what you ignore. Then it shows you more of what keeps you there. It is designed for attention, not for truth or quality. Kids need to understand this, otherwise the feed trains them.

AI works with probability, not certainty. It predicts the most likely answer based on patterns. It can be useful, but it can also be wrong, and it can sound confident while being wrong. That is why the key lesson is verification: ask for sources, cross-check, compare.

Money shapes content too. Ads and sponsored posts can push messages higher. And promoted influence will increasingly affect AI answers as well. If we do not teach people to recognize paid visibility and to trace information back to reliable sources, they will be guided by whoever pays.

A phone is not only distraction. It can be a learning tool: look up the meaning of a word, check a date, understand a concept, satisfy a curiosity fast. AI can support studying by asking targeted questions, helping you spot gaps, and explaining the same topic in different ways. Support, not replacement.

Bans are easy. Education works. We did not ban electricity because it is dangerous. We taught rules and built habits. Digital tools need the same: guidance, culture, and responsibility.

#ArtificialDecisions #MCC

First they tracked us. Now they build our decision twin.

Today there is something very real: a decision twin. It is a digital model of a person, built using data that already exists. Nobody needs to follow us live. They just query the model.

Companies collect data from many sources: web browsing, apps, location history, online and offline purchases, loyalty cards, streaming habits, social activity. Everything ends up in structured databases. These databases are analyzed with AI systems, often internal, often local. The system looks for patterns.

When someone types “John Doe”, the system returns a behavioral profile: daily routines, places visited, spending habits, price sensitivity, content watched, reaction to offers, purchase probability, risk level, priority inside automated systems.

Each piece contributes. The loyalty card adds purchases. The smart TV, through ACR, adds viewing habits. The phone adds movement and time. The browser adds attention and reading behavior. The car adds driving style. Wi-Fi adds presence and duration.

This decision twin is used to choose how to interact with that person: prices, ads, content, offers, access to services. Decisions happen before the person is aware.

This is driven by market logic. Companies compete on how well they can predict people. Better prediction means more value. There is no full exit. Leaving the digital world is not realistic. There is only damage reduction.

Every time we give data, we choose who gets it. Every app is a source. Every loyalty card is a trade. Every permission is a transfer. Limit permissions. Connect fewer accounts. Choose tools that collect less data. Treat data like money.

The decision twin grows by accumulation. Reducing that accumulation is the only real control we have today.

#ArtificialDecisions #MCC

The Chatbot That Moves Your Vote

Here in the United States, political persuasion is changing shape. It now happens through private, one-to-one conversations, with an artificial voice that answers, adapts, and pushes. A few minutes of dialogue can change an opinion.

A TV ad speaks to everyone in the same way. A chatbot follows the topics of the person in front of it: jobs, taxes, healthcare, cost of living. Every objection gets a reply. Every hesitation gets attention. The conversation becomes targeted, persistent, personal. Chatbots persuade more when they fill answers with data, references, examples.

Now add the next step. Chatbot answers will become advertising formats, like sponsored posts on social media. Someone will pay to push a reply, a source, a topic into the conversation. The message arrives with the neutral voice of an assistant that seems helpful, while it guides choices.

All of this happens in private. Each person receives a different version. No one sees what others are told. When political opinions are shaped inside personalized, monetized chats, power moves toward those who control technology, data, and budgets.

#ArtificialDecisions #MCC