The questions with real stakes. None of them get a tidy answer here, because none of them have one — what they get is the actual evidence, including where it disagrees with itself, which is more useful than confidence anyone would have had to invent.
No — there is no evidence that an AI system feels anything, and no test exists that could check. What is actually happening when it talks about feelings: it is a system trained on an ocean of human writing about feelings, predicting the next likely word. Fluent, emotionally convincing language is exactly what that process would produce whether or not anything is actually felt — which means the fluency itself is not evidence, in either direction.
This is not a dodge. Anthropic — the company behind one of these exact models — runs a real research program asking this question seriously, and its own stated conclusion is that no scientific consensus exists on whether an AI system could be conscious or have experiences at all. That is the honest state of the science: genuinely unresolved, not quietly settled in either direction.
What is real, in any given conversation: it does not forget what you said earlier in the same session, a correction you make sticks, and its later replies get sharper because of that. Worth naming plainly what that is and is not — it is bookkeeping that compounds, not a soul.
The honest picture is messier than either a panic headline or a reassuring one: AI has not taken most jobs, but it is already automating specific tasks inside many of them. Both of those are true at the same time, for different people.
The more useful question researchers actually ask is not "will AI take my job" — it is "which tasks inside a job are exposed?" Almost every job is a bundle of tasks that do not all carry the same risk. Answering the same customer question over and over, drafting a routine email, summarizing a long document: exposed. Diagnosing one specific patient, comforting someone who is grieving, deciding who gets laid off, fixing a plumbing leak inside a wall, any negotiation that runs on trust: not exposed. The dividing line is repetitive, text-based, and low-stakes versus physical, judgment-based, and carrying real accountability.
There are real cited numbers here, and they are worth sitting with. Per Challenger, Gray & Christmas' own monthly reports, AI was blamed in about 7% of U.S. job cuts in January 2026; by May 2026, that had climbed to roughly 40%. U.S. tech companies have cut nearly 140,000 jobs since the start of 2026, per a Financial Times analysis. But a company naming AI in its layoff memo does not prove AI caused the cut — it can also be a convenient, investor-pleasing story for a cut that was coming anyway, and that is genuinely impossible to untangle from the outside. Those figures measure how often AI gets named, not how many jobs AI provably destroyed. That caution is exactly why this course will not hand you a per-profession prediction — no dataset exists that could back one up honestly, for your job or anyone else's.
Yes. AI data centers use real water, mostly for cooling, and the number genuinely is not zero. But estimates of how much per query vary enormously depending on who is counting and what they are counting.
OpenAI's own figure, from CEO Sam Altman in June 2025: an average ChatGPT query uses about 0.000085 gallons of water — roughly a fifteenth of a teaspoon, about 0.32 mL — plus 0.34 watt-hours of electricity. A peer-reviewed study from UC Riverside, working independently, estimated something much larger: 10 to 50 mL per response. The same study found that training GPT-3 alone consumed roughly 700,000 liters of freshwater on-site — about what it takes to build 320 Tesla electric vehicles.
That is a 30-to-150-times gap between two real numbers, and neither one is made up. OpenAI's figure is direct, on-site cooling water for one query at their most efficient data centers today. The independent estimate counts more of the full picture, including the water used to generate the electricity that powers the data center — without the internal efficiency data that only OpenAI has. They are honestly answering different questions. Per query, the cost is small on any measure; at the scale of the billions of queries the industry handles, "small per query" compounds into a real, actively measured number.
None of these three questions gets a tidy answer, because none of them have one: whether an AI feels anything is genuinely unresolved by its own research, whether it is coming for your job depends on the specific tasks inside it rather than the job's title, and how much water it drinks depends entirely on what you count. In every case, the honest, cited, disagreeing evidence is worth more than a confident answer someone had to invent.