Why I’m taking AI seriously

AI seems to attract two reactions that make it difficult to discuss: dismissing what it can do, or fearing what might happen if it works.

There is something familiar about that. The internet changed how people worked, learned and communicated, and AI raises another set of questions about what change could mean. A familiar pattern does not settle whether a particular concern is justified. It does make immediate dismissal feel premature.

My view is that AI is still widely misunderstood. Too much of the discussion seems to stop at the costs or the frightening possibilities, with less attention to the useful things ordinary people might gain.

That is the starting point here. This is an amateur investor’s and AI user’s view of the technology, the evidence worth considering and the uncertainties that remain.

What using AI has made possible

Some of the reasons for taking AI seriously come from using it on real projects.

Arxonic itself is one example. AI has helped turn the idea for this publication into a website, including a directory where readers can learn more about the people mentioned in the research. Decisions about what belongs here, how it should look and what it should say have developed through questions, drafts and revisions.

Another is Outrank, a small website built around an idea: people bid for a position on a board to promote a product or profile. AI has helped with the build, including the ranking system and payment integration. Building it does not establish whether people will want it. It gives the idea something concrete to test.

The work also extends beyond websites. Research can be organised into a library where claims stay attached to their sources, disagreements can be compared and earlier conclusions can be revisited. That is useful when trying to understand a subject as broad and changeable as AI. Each new piece of information has somewhere to fit.

What matters to me is the range of work this makes possible. An idea can become a working project, and a question can become something that is investigated properly. There are still choices to make, errors to catch and results to check. That has been part of building and refining these projects.

The possibilities go beyond software

The broader argument is about efficiency and access to help.

If AI can reduce the effort needed to complete useful work, there is a route to lower costs. Whether those savings reach customers depends on how products and services are priced, but the possibility matters.

The same thinking applies to research. Instead of exploring only a handful of promising approaches, AI could help researchers investigate more possibilities and identify which deserve further testing. Faster development of medicines is one of the outcomes that makes this worth taking seriously.

There are concrete steps in that direction. A study published in Nature Medicine in June 2025 reported a phase 2a trial of rentosertib, a drug candidate for a lung disease called idiopathic pulmonary fibrosis. AI was used in identifying the target and designing the molecule. It was an early clinical trial, not proof of a cure or a guarantee that drug development will become faster across the board. It does show the work reaching clinical testing. The study.

Learning is another possibility. Help that adapts to the learner could make it easier to revisit a difficult explanation, work at a suitable pace and build new skills. In a randomised study published in 2025, a carefully designed AI tutor produced greater learning gains in less time than an active-learning class for the undergraduate physics material tested. That is a result for a particular tutor and setting, rather than every chatbot or subject. The study.

I agree with taking these possibilities seriously when discussing AI’s costs. Personal tools, more accessible learning and progress in research are all part of what the technology could make possible.

Putting the water debate in context

Data-centre water use is one area where the discussion needs more proportion.

Golf is a useful comparison because it is another activity that uses water and provides a benefit to the people using it. The question is how consistently those costs and benefits are weighed.

The US golf industry’s national survey estimated that facilities applied about 531 billion US gallons of irrigation water in 2024. Berkeley Lab estimated that US data centres directly consumed about 17.4 billion US gallons in 2023. Golf survey, Berkeley Lab report.

Those are different years and definitions: water applied to golf courses versus water consumed directly by data centres. The data-centre number also excludes consumption associated with generating their electricity, which Berkeley Lab estimated separately at nearly 800 billion litres (about 211 billion US gallons). These figures cover data centres overall, not AI alone. They provide context, not a clean ratio of environmental harm.

Location matters too. The Oxford-hosted research discussion explains that water use in an area with plentiful supply can have different consequences from use in a drought-stressed region. Electricity generation also needs to be included when considering the wider footprint. Oxford’s explanation.

Golf faces criticism over water as well; its own industry body acknowledges that. The comparison cannot establish that nobody objects to golf. GCSAA’s water overview.

My concern is with treating a large water-use figure as enough reason to dismiss AI as a whole. The potential benefits reach well beyond a single activity or group of users. That deserves a place in the discussion alongside the local costs and the choices about where and how facilities are built.

Demand needs somewhere to run

Those facilities connect the wider AI story to compute: the processing resources used to train and run models.

There seems to be more demand for compute than can easily be supplied, especially if data-centre construction falls behind. That is part of the reasoning behind seeing access to compute as valuable.

There is evidence of constraints. The IEA’s 2026 assessment describes bottlenecks across energy equipment, chips and grid connections, despite large investments and a growing pipeline of projects. It also notes that investment to relieve those constraints could allow more capacity later. IEA assessment.

That supports concern about supply keeping up. It does not establish a permanent shortage of every kind of compute. The type of hardware, location, electricity supply and the service someone needs all matter.

The two possibilities worth following are demand growing faster than capacity, and better AI making the available capacity more useful. Both could affect the value of access, but the outcome is uncertain.

Why there is money behind this view

Direct investment in private companies such as OpenAI and Anthropic has not been an accessible route for me as a non-accredited investor. My connection to Venice began through using it in 2024, and the assets came through that involvement.

Today, DIEM and VVV are part of my exposure to AI. Keeping them is a decision to remain exposed to something whose usefulness and future value are both part of the appeal.

DIEM makes that connection tangible. Under Venice’s current published model, one staked DIEM provides $1 of API credit that renews each day. VVV is linked through Venice’s staking and DIEM-minting system. These are Venice ecosystem assets with their own economics; they do not give ownership of OpenAI or Anthropic. Venice’s explanation.

The recurring credit is a practical use for the asset. Its future market value is a separate possibility. A dollar of credit is an allowance to use a service, rather than a fixed quantity of hardware or guaranteed availability of any particular model.

That leaves a gap between the wider compute argument and an investment result. If inference prices rise, the allowance could buy less work. If capability improves or prices fall, it could buy more useful work. Venice’s delivery, pricing, token supply and demand for its services all affect the connection.

I expect improving AI could make the compute more useful, but I don’t know how far that will go or what it will mean for the assets’ prices. That uncertainty is part of the position.

A starting point to return to

The view so far is that AI deserves more serious consideration than outright dismissal. The work on websites and research already gives it practical value for me. The possibilities for efficiency, learning and research give reasons to investigate further.

There is also a real stake in understanding where this goes. Holding DIEM and VVV makes the reasoning consequential, without making it correct by default.

These essays will be a record of that reasoning as it develops: what the evidence supports, where there is disagreement, and what remains unknown. The aim is for readers to understand the thought process and have enough information to form their own view.