Category: Artificial Decisions

Smart Robots Will Be Everywhere Like Smartphones: Here’s Who Will Run the Next Machine Era

You’ll see them at home, in factories, in warehouses, in stores. Like phones today, they’ll feel normal. They’ll be as smart as the models we use to write and answer. This time they’ll have a body: hands, strength, balance. Walk the dog, clean the floor, fix a drawer, swap a broken part, cover a night shift on a line.

They’ll learn us, our routines, our preferences, our spaces.

Here in the U.S., real tests are already happening. BMW has publicly tested Figure 02. Hyundai has controlled Boston Dynamics since 2021. NVIDIA is building the compute and simulation stack for humanoid training. Apptronik and Agility are pushing robots into real warehouse operations.

The leaders are taking shape: Tesla for scale ambitions, Figure for general-purpose factory work, Boston Dynamics for mobility, Apptronik for logistics, Agility for distribution centers, NVIDIA for the underlying infrastructure.

The trigger is price. When a robot costs roughly a year of wages, it becomes a straightforward business decision. Jobs will shift fast, some roles will grow, others will shrink. Policy needs to move before this becomes a social emergency.

What do you think?

#ArtificialDecisions #MCC

What if AI Won’t Replace People, but Companies?

Will AI replace us because it can do office work better than humans? That’s the question everyone is asking. My answer stays the same: the first to be replaced will be the ones who don’t use Artificial Intelligence. I’m convinced.

Now push it one step further. What if this applies to companies too? The first companies to be replaced will be the ones that don’t adopt AI, replaced by firms that deliver the same service faster, cheaper, with AI at the core.

Past industrial revolutions weren’t just about machines replacing jobs. They were about productivity: more output with less input. AI follows the same path, with one key difference: it operates on language, and language is the raw material of entire industries.

When companies adopt AI to stay competitive, they expose their industry’s “grammar” to the system. Every prompt, every revised document, every client response becomes training data about how that business thinks, writes, argues, and manages risk. At scale, AI doesn’t learn one company. It learns the patterns of the whole ecosystem, then can reproduce them in an optimized way.

Think law, consulting, finance. AI absorbs contract structures, due-diligence patterns, standard answers. Over time it can deliver parts of those services directly to the end client, without the full organizational layer in the middle. The risk isn’t only the employee. It’s the company as an intermediary.

AI still needs competent human supervision. It’s probabilistic. It can be wrong in subtle ways. The advantage isn’t “using AI.” The advantage is knowing its limits, controlling its output, deciding what to delegate and what must stay under human responsibility.

That’s why I keep saying it: first go the people who don’t use AI. Then the companies that don’t use AI.

#ArtificialDecisions #MCC

How AI Works, in a Simple Way

I’ll explain, simply, how LLMs “think.” Stay with me. It’s dense, but this is the easiest version that still stays accurate.

We write a prompt. One sentence. The model breaks it into tiny pieces called tokens. Sometimes a token is a full word. Often it’s part of a word. That matters because it decides how much text fits in the context, and it changes by language. Then each token becomes numbers. Lots of numbers. Coordinates in a huge space. For an LLM, text is vectors. And it runs repeated operations on those vectors, mainly matrix multiplications. Computation. Just computation.

Next, the tokens get linked to each other. Some weigh more, some less. The model assigns numeric weights across the text, based on what’s in front of it right now. It works. That’s why it fools us. It looks like understanding. After a few passes, it does the key step: it produces probabilities for the next token. It picks one, appends it, recalculates, and repeats. One token at a time. Dozens, hundreds of times.

The “intelligence” feeling comes from continuity. Correct grammar, consistent tone, smooth flow. But the engine is prediction. If a continuation sounds plausible because it matches patterns it has seen, it may choose it even when it’s wrong.

So yes, it can write perfect sentences with incorrect content. If we don’t give strong constraints or reliable documents, it fills gaps with what sounds best.

There’s also a setting called temperature. Low temperature means safer, more predictable choices. Higher means more variation.

When we ask “how does it know?”, often it doesn’t. It has seen similar patterns. It has learned how sentences usually continue on that topic. And the more we use it for money, health, contracts, or reputation, the more we should remember what it is: a machine that predicts words.

#ArtificialDecisions #MCC

Sometimes Delegating to AI Causes Damage. Sometimes It’s Extremely Useful and Saves Time. But When Should You Use It, and When Shouldn’t You?

The decision is simple. It comes down to three things you have to consider together.

First: Human Baseline Time. The real time it takes you to do the task yourself. If a tricky email takes you 5 minutes, Artificial Intelligence often isn’t worth it, because you’ll spend more time prompting and fixing than writing. If a report would take you two hours, AI can be a real advantage.

Second: Probability of Success. The chance the AI gives you something good enough on the first try. Summaries, first drafts, and translations are usually high. Legal, medical, or strategic calls are usually low, even when the answer sounds confident.

Third: AI Process Time. The time you spend asking, waiting, reading, checking, correcting, and redoing. If that process time matches or beats your human time, delegation doesn’t pay.

AI delegation works when human time is high, success probability is high, and AI process time is low. If one of these breaks, AI stops being an accelerator and becomes a brake.

One example: a standard blog post. By hand, 45 minutes. AI can give a usable draft quickly. Worth delegating. Another example: a sensitive reply to an angry customer. By hand, 10 minutes. AI often misses the tone, and your total time becomes 20. Not worth it.

One counterintuitive truth: the more expert you are, the more useful AI becomes. Experts give better instructions and spot errors fast. Non-experts spend too long figuring out whether the output is right, and risk goes up.

AI is a speed multiplier, not a substitute for judgment. This isn’t ideology. It’s a calculation: time, probability, and control.

#ArtificialDecisions #MCC

Attention. Today Online It’s Easy to Make Us Believe an Idea Is Supported by “Everyone,” Even When It Isn’t

Thousands of comments cheering it on. Thousands of likes. You see that and you adjust your opinion. Now that kind of consensus is easy to fake, because it’s often not people talking. It’s networks of autonomous Artificial Intelligence agents writing, replying, arguing, applauding, attacking, 24/7.

Consensus becomes something you can copy. You read the same comment a hundred times, you see a thousand likes, the mood feels set. Some people follow it. Some stay silent. Some get angry.

Tools anyone can use can flood social platforms and forums with credible profiles: photos, stories, natural language, human mistakes, jokes, rage, even warmth. And they can push thousands of posts. No need for one big “media lie” like the old days. Small phrases, same direction, are enough.

Whoever controls these networks can steer them for or against anything. One message to one group, a different one to another group. Manipulation becomes scalable, personalized, and invisible. Finding who’s behind it is hard: campaigns spread across thousands of nodes and platforms slow everything down.

We still judge individual people by reputation, while what we need are global rules: traceability for coordinated campaigns, real transparency obligations for anyone using networks of autonomous agents to influence public opinion and democratic processes. And the platforms? They don’t look in a hurry.

What do you think?

#ArtificialDecisions #MCC

The Whole Truth About AI and Water Consumption

In Texas, in 2025, data centers used about 90 billion liters of water, including cooling and water linked to electricity production. By 2030, estimates reach up to 600 billion liters per year.

In just a few years, Artificial Intelligence is competing with agriculture, cities, and industry. Computing systems generate heat. Heat must be removed. Some data centers can use up to 1.5 million liters of water per day just for cooling.

You might think: water evaporates and comes back as rain. So what is the problem?

Water returns, but not where it is needed and not when it is needed. A data center draws fresh water from a local basin. Part of it evaporates and leaves that territory. It may fall elsewhere, even over the ocean, or months later. Meanwhile, the local community has less water available, often during the hottest weeks, when demand is already high.

There is another issue. Demand grows fast. Water recharge is slow. Aquifers are not infinite. If withdrawals exceed natural recharge, water levels drop, wells go deeper, costs rise, drought risks increase. A global water cycle does not fix a local shortage created in a few years.

There is also indirect water use. Electricity production often consumes water. That indirect share can represent up to 75% of the total footprint. The data center reports one number. The energy-producing region carries another, often larger, burden.

AI infrastructure expands where land and energy are cheaper and permits move faster. These areas are often already exposed to heat and water stress. Multiply that pattern across regions and the pressure on local water systems increases.

Water does not disappear. Availability, in the right place at the right time, becomes scarcer. What do you think?

#ArtificialDecisions #MCC

Warning, They Can Know Where You Are in Real Time. How to Protect Yourself

All it takes is an AirTag placed in your bag for anyone to know where you are, in real time, without you noticing. This part is well known. iPhones alert you automatically if an AirTag that is not yours is following you. But if you have an Android phone, nobody warns you. Here is what you need to do to stop anyone from spying on your location.

Roughly 70 percent of smartphones run Android. Most people are exposed and have no idea.

AirTags work through a network of nearly two billion Apple devices worldwide. Every iPhone that passes near an AirTag silently updates its position, even if that iPhone is not yours. Someone just needs to walk past you. Great for finding lost keys, but unfortunately also great for following a person without them knowing.

On iPhone protection is automatic. If an AirTag that is not yours follows you, you get a notification, no setup needed: “Watch out, there is an AirTag that is not yours following you.”

On Android alerts exist, but they only work if you do not have an old phone and if you have Google Play Services updated. If you do not, nobody warns you. And even when it works, independent tests show that detection is slower and less reliable. You could be followed for hours before anything warns you. If it warns you at all.

To help protect you, AirTags have a small speaker that makes a sound after several hours if separated from their owner. A signal to notice something strange in your bag. The problem is that a small hole drilled under the battery disconnects the speaker. No more sound. No more warning. And these modified AirTags are already being sold ready to use on eBay.

How to protect yourself: if you have Android, go into settings and search for “unknown tracker alerts.” Make sure it is on. Download AirGuard, free, open source, built by a German university, no commercial interest. It scans in the background and alerts you if something is following you.

Apple and Google are working together on a shared standard called DULT for detecting unwanted trackers. The direction is right. But the problem exists right now.

iPhone users are mostly covered. Android users: settings, “unknown tracker alerts,” turn it on. And install AirGuard.

#ArtificialDecisions #MCC

Why AI Hallucinations Are Causing Real Damage and We Don’t Notice

Artificial Intelligence hallucinations cause damage because they end up inside decisions. We ask a question, we get an answer that looks clean, organized, confident. We read it fast, paste it into a document, forward it. The damage starts there.

The mistake is usually in the details, and details are what people check the least: a number with the comma in the wrong place, a date, an agency name, a deadline, a rule that is real but applied to the wrong place. The text still sounds credible, so the error stays.

At work it happens like this. We ask for market growth, user counts, a percentage to use. We get a precise-looking number with a neat explanation. It goes into a report, then a slide. Nobody opens the original source because the answer feels ready. If the number is invented or just wrong, budgets and priorities go off track, and we notice later.

At home it’s the same, with higher stakes. We ask about symptoms, medicines, dosages, interactions. The answer is calm and structured. One wrong detail can change what someone does.

The biggest risk is the chain. A false line becomes the base for the next question. We paste it into a new prompt and the AI builds on it. The next answer feels even stronger because it has more context, but it is strengthening the first mistake.

Protection is simple. Any AI answer that contains facts stays a draft until we verify a primary source ourselves: the original document, an official page, the full text. Check details first: names, dates, numbers, quotes, deadlines. When money, health, contracts, or identity are involved, decisions wait for an external check.

AI can write fast, but verification stays on us.

#ArtificialDecisions #MCC

Big Tech Is Laying People Off Because of AI. What to Do Now to Keep Your Job

In 2025, the tech sector recorded 122,549 layoffs across 257 companies. HP plans 6,000 job cuts by 2028 to move resources into Artificial Intelligence. Meta cut about 3,600 roles, Microsoft 6,000, Amazon 14,000 corporate jobs. Office work gets reduced, AI-related roles get funded. Follow me to the end, because the fix is practical.

The people leaving and the people being hired are not the same. Companies cut back office, admin, content moderation, customer support. They hire ML engineers, researchers, data and security specialists. Moving from one side to the other takes time.

Goldman Sachs Research estimates generative AI could reduce 6–7% of US jobs. The roles most exposed include programmers, accountants, legal assistants, customer service, and credit analysts. Tasks like writing, summarizing, classifying, and answering can be automated.

So what do we do? Use AI seriously, every day, on your real work. Put it into documents, emails, processes, numbers. Track time saved and quality improved. Save examples and results so you can prove what you can do.

Become the person who runs the work with AI: sets context, checks output, verifies, signs off, takes responsibility. Push your company for real training, tied to actual roles, with clear rules and a practical plan for the next 6–12 months.

Tell your CEO to bring in experts and make decisions now: map where AI is already used, decide what can be sped up and what must stay human, invest in skills and tools, assign clear ownership. Tag your CEO or your manager.

#ArtificialDecisions #MCC

Altman Defends AI: “Raising a Child Costs Energy Too”

Sam Altman has found a way to answer criticism about Artificial Intelligence’s environmental impact: compare it to humans. Speaking at an Indian Express event during an AI summit in India, the OpenAI CEO said that training a human being for twenty years costs a lot, so the comparison with ChatGPT is fairer than it looks.

But come on. A human being lives, works, creates, reproduces, contributes to society in ways no probabilistic system can replicate. Reducing a human life to a training energy cost is a rhetorical move, not a scientific comparison.

On water, Altman called the 64-liter-per-query figure completely false, with no connection to reality. Google published its own numbers for Gemini queries: 0.24 watt-hours of energy and 0.26 milliliters of water. The 64-liter estimate came from a University of California study that included not just data center water but also the water used by the power plants supplying the servers, a methodology widely disputed, and the only public reference available, because companies are not required to disclose their own figures.

The energy impact is real. According to the IEA, global data centers consumed around 415 terawatt-hours, about 1.5% of global electricity. Projections point to nearly double by 2030.

There are no legal obligations forcing tech companies to disclose how much energy and water their data centers consume. Independent researchers build estimates from the outside. Companies can deny any figure without having to provide their own. When Altman says the 64 liters are false, he may be right. But he is asking us to trust his word on data he is not required to make public.

Data centers keep pushing up electricity prices in surrounding areas. A concrete, measurable impact that no evolutionary metaphor can fix. What do you think?

#ArtificialDecisions #MCC

5 Things to Watch When You Bring AI Into a Company

First, Artificial Intelligence often shows up before the company officially adopts it. Employees use it quietly, and many use it the wrong way. A report by the National Cybersecurity Alliance and CybSafe, cited by Security Management, says 43% shared sensitive work information with AI tools without their company knowing. That includes internal documents, financial data, and customer data. Cisco’s 2025 Data Privacy Benchmark Study says almost half of respondents admit they entered employee personal data or other non-public company data into generative AI tools.

Second, a prompt is a packet of company information. When you paste an email, an Excel file, a contract section, or a sales note, that content leaves the company perimeter and goes to systems you do not control. You usually do not know where it ends up, how long it stays, or who can use it later.

Third, clean the data before you use AI. Remove anything that points to real people, real clients, real products, real numbers. Replace names with placeholders, use “find and replace,” and use codes instead of identities. Keep a local mapping so you can restore the original names after you get the output.

Fourth, be careful when you connect AI to company documents or to the web. AI reads text and tends to follow instructions inside text. Attackers can hide malicious instructions inside a file that looks normal, pushing the AI to search for things it should not touch, including internal folders. Keep access minimal, enable connectors only when needed, and stop immediately if the AI behaves oddly.

Fifth, responsibility stays human. AI can sound confident while being wrong. For anything involving money, contracts, people decisions, hiring, penalties, or deadlines, require human review. Keep a short internal policy with clear examples of what can be pasted and what cannot, and require company accounts and approved tools.

#ArtificialDecisions #MCC

Free AI Is a Fiction, and It’s Often Ripping You Off

We grew up thinking that everything on the internet is free. Search engines, social networks, digital services. We learned that digital tools come without a real price. Today, especially with Artificial Intelligence tools, that idea no longer works. Free is an illusion.

When we use a free AI, we are not really using the tool. We are seeing a reduced, limited version. It exists to show that the product is there, not what it can truly do. Then people get disappointed: it makes mistakes, feels shallow, looks useless. From there comes the idea that AI does not work. What we are judging is a weakened demo.

Digital tools are no longer accessories, they are daily life. Work, study, communication, money, decisions. Technologies do not go backwards. The internet stayed. Smartphones stayed. Social networks stayed. AI is following the same path, faster and with deeper impact.

For years, advertising paid for everything. Today it is not enough. AI is expensive: computing power, energy, data centers, security, constant updates. Running advanced models costs a lot, every single day.

If you do not pay with money, you pay in other ways: data, behavior, attention, time. This logic already exists in the digital world. With AI, it becomes even clearer.

We already pay subscriptions for music, movies, cloud services, software. Expecting AI to be truly free is just a habit from the past, not an economic reality.

When people say “AI always makes mistakes” or “I tried it and it is useless,” they are usually talking about free versions. There is no real comparison with paid ones: depth, continuity, reasoning, reliability all change.

Open source exists, but it requires real skills: time, study, technical ability. It works for experts, not for everyone. For most people and companies, tools that truly work have a cost. This is a fact. The age of everything for free is over. Pretending otherwise only means staying behind, while others are already using AI seriously.

#ArtificialDecisions #MCC

Artificial Intelligence That “Challenges God”? I Don’t Buy It

AGI means Artificial General Intelligence. The promise is a machine that can handle any mental task like a human. Learn new things without being reprogrammed, move from medicine to math to strategy with the same “brain.” Use context, experience, common sense. Grow like a human mind. That’s why people talk about it in almost religious terms.

What we have today is different. Even when it looks flexible, it’s still a specialist tool. A language model writes, summarizes, answers, and codes because it learned patterns of language, not because it understands the world. Each reply is a probability guess: which words are most likely next, based on its training data. It can work well inside its comfort zone. Outside that zone, it becomes fragile.

A real AGI would be stable and consistent over time. Today’s systems are not. Each output is a one-off: no personal experience, no awareness of mistakes, no intention. When it’s wrong, it doesn’t know it’s wrong. The “mind” feeling comes from good language, not real understanding.

Real autonomy also isn’t here. An AGI would decide what matters and how to learn. Today’s AI depends on human-chosen data, human goals, and human metrics. The fence stays up.

I don’t believe in the AGI hype. Meanwhile, today’s AI is already inside decisions that matter: hiring, money, health, contracts. Treating it like a “general mind” makes people delegate too much. Treat it as what it is: a powerful statistical tool that still needs human control.

#ArtificialDecisions #MCC

Watch Out for New AI Scams

Here in the United States, people write to me every day. “My son called me.” “I saw my boss on video.” “The email looked perfect.” I spoke with a senior FBI official in New York who said generative AI is making scams more believable and easier to scale.

Voice cloning needs very little. A short voice note or a public video is enough. Then comes the urgent call: accident, lawyer, bail, broken phone. The goal is speed. If you rush, you pay. Video deepfakes raise the trust level even more. A face on screen still feels authoritative, even if it can be faked.

AI also improves phishing emails and fake customer support chatbots. They look clean, reply fast, and ask for “verification.” That means OTP codes and access to your email, then accounts disappear. Chats, calls, audio, and video are data. If money or access is involved, verify outside the channel. Hang up and call a saved number. Open the app, not the link. Never share a code from SMS or an app.

#ArtificialDecisions #MCC

I’ll update you on how things went in Berlin and on how we want to create a World Council for AI, together with a decalogue of “commandments.” They may sound like technical topics, but they mainly concern society and everyone’s everyday life.

🇺🇸🇬🇧 Clip: The World Forum

Today, I spoke about several facts that are obvious to those working in the industry, but which are often not so for the general public. I recalled that large AI platforms respond to economic incentives that do not always coincide with democratic resilience. Engagement generates revenue. Growth generates valuations. This influences design choices.

We have seen how easy it is to manipulate perception. The AI-generated video of the “Albanian Minister” was perceived by some as neutral evidence, almost incorruptible. In reality, behind every output, there is a model trained by someone, with choices regarding data, filters, and objectives. And that infrastructure can be attacked, altered, or compromised. The question is not whether a content is convincing, but who designed it, who controls it, and who can intervene on it.

I emphasized that ethics is often treated as communication or compliance, whereas it should be a structural constraint. A single, generic ethical framework is not enough for systems operating in different cultural and legal contexts. We need contextual, configurable, and verifiable ethical parameters.

Then there is a geopolitical dimension. States regulate at a national level, but systems operate on a global scale. Offshore hosting, VPNs, and multiple jurisdictions make the application of regulations fragmented. Without international coordination, regulation remains fragile.

Even when talking about open source or open weights, transparency regarding training data remains limited. We do not know for certain what has been included, excluded, licensed, or acquired. And in local models, guardrails can be easily removed.

Meanwhile, daily behavior is changing. More and more people, even very young ones, are turning directly to an AI system to inform themselves or make decisions.

The central point concerns artificial decisions. As long as a system suggests, the human being formally maintains control. When a system passes from assisting to acting, the risk profile changes. Decisions are delegated, along with power, responsibility, and risk.

In healthcare, systems already exist that contribute to establishing priorities for interventions and access to care. In the financial sector, they evaluate mortgage grants and economic conditions. With domestic robots and autonomous systems, these choices can extend to physical safety. If a robot detects a fire in a house with an isolated elderly person, which action does it prioritize? On what criteria? On what ethical rules incorporated into the model?

These systems appear objective because they speak in a fluid and coherent way. We tend to trust what reasons like us. But they operate invisibly, within rankings, moderation systems, recommendation engines, and conversational interfaces.

A single probabilistic error can seem irrelevant. Millions of decisions delegated every day redefine access to information, reputations, and opportunities. A security problem, a contamination in the data or in the model, can propagate on a large scale before we even notice.

For this reason, I proposed that the World Council focus on high-impact decision-making systems: real independence from governments and platforms, transparency on funding, mandatory audits with public and repeatable criteria, and clear responsibility in case of systemic damage.

The concrete proposal is an independent certification dedicated to high-impact artificial decision-making systems, with repeatable public tests, mandatory real-time reporting of incidents, and defined escalation procedures. Certification as a condition for participating in the public space and for obtaining trust.

In Ten Years, Work Will Look Different. Here’s What Still Makes Sense to Study Today.

My son is 13. University in five years. A degree in ten. In between, AI will write more code, and robots will enter hospitals and care facilities. The question is simple: what should a kid study today? Classics or computer engineering?

Start from passions, always. But some degrees are more resilient than others. Healthcare with technology inside: robots will take physical and logistical tasks, but people will still manage clinical data, protocols, and risk. Nursing, health professions, biomedical engineering, clinical informatics, biostatistics, healthcare management. Data and decisions: statistics, data science, applied math, econometrics, operations research. AI speeds up analysis, and it speeds up mistakes without method. We need people who can measure, validate, and understand causality.

Security and digital infrastructure: cybersecurity, networks, cloud, industrial security. More automation means more exposure. Hospitals, schools, companies, cities need people who protect systems and respond when they fail. Physical-world engineering: electrical, energy, mechanical, industrial, materials, automation, supply chain. Robots are tools. They increase the need for design, maintenance, certification, reliability. Technical law and governance: privacy, liability, rules for automated systems, tech contracts, IP, labor. We need specialists who understand both law and systems.

Everything else can work if done seriously. Creativity and tourism often work best as a second axis. Tools can be learned anywhere. YouTube teaches tools. University should teach method, numbers, responsibility.

#ArtificialDecisions #MCC

Creative AI: Control Matters More than Power.

Here in the United States I see it when I talk with designers, filmmakers, and creators. Everyone uses AI. Many are frustrated for one clear reason: AI can generate fast, but real creative work needs control and repeatability.

In real work, creativity is a chain of choices. First idea. Variations. Selection. Edits. More edits. Consistent style. Final delivery. The hard part is not getting “something nice”. The hard part is getting the same thing again, improved, without losing identity.

Many tools still work like a slot machine. You type a prompt, you get a good image or video. Then you ask normal production requests: same character, same lighting, same style, change only the camera angle. Same layout, change only the text. Same scene, change only one object. And everything changes.

Simple example: a school event poster. AI gives you ten nice options in one minute. Then parents ask for real changes: new time, bigger logo, Instagram size, print size, English version. If every small change breaks the style, you lose time after, not before.

That is the point: power makes many outputs. Control keeps continuity. Control means consistent identity, small edits without breaking everything, and the ability to deliver a repeatable system. The rule of the video is one: use AI inside a controlled workflow. Clear references. Clear steps. Comparable versions. Edits that keep the same look. Repeatable delivery.

#ArtificialDecisions #MCC