Cheaper AI won’t mean less AI — and human judgment is the real price
Cheaper AI won’t mean less AI — and human judgment is the real price
Will Cheaper AI Automate Everything?
The Jevons paradox, collapsing inference costs, and why making AI cheaper may create more work for people, not less.
(15:00–16:00 Tbilisi)
Avtandil Gagnidze, Professor and Vice-President of East-West University, joins the panel of the AI for Developing Countries Forum (AIFOD) Weekly AI Review.
Speakers
- Valerie Pelton — General Counsel, Tracked Biotechnologies LLC (United States)
- Dr Craig Gibbs — Executive Manager, JET Education Services (South Africa)
- Fabrizio Degni — Chief AI Officer, Webuild (Italy)
- Reda Sadki — Executive Director, The Geneva Learning Foundation (Switzerland)
- Avtandil Gagnidze — Professor and Vice-President, East-West University (Georgia)
Background
In September 2026, TypeSafe AI released Jev, a model built to make structured judgments rather than to generate text. Jev does not write reports or explain its reasoning: given information, it returns a choice, a score or a probability that software can act on directly — classifying documents, ranking options, selecting tools, filtering content or deciding whether a case needs human review. Its confidence scores allow software to proceed automatically when confidence is high.
The vendor reports a cost of about US$0.042 per million input tokens and response times of 70–500 milliseconds. These figures are not independently verified, but they point to a clear trend: AI judgment is becoming faster and far cheaper. If costs keep falling, organizations may embed AI evaluators in almost every digital workflow.
What is the Jevons paradox?
Jev takes its name from the British economist William Stanley Jevons. In The Coal Question (1865), he observed that more efficient steam engines did not reduce coal consumption. Efficiency made steam power cheaper and useful in more industries, and as adoption spread, total coal use rose.
Greater efficiency can lower the cost of using a resource, increase demand for it and ultimately raise total consumption.
The same logic may apply to AI:
Cheaper AI tasks → more AI applications → greater total AI use
A more efficient model may use fewer resources per task, yet lower costs encourage wider and more frequent use, including in activities that were once too expensive to automate. The cost of each operation falls while total demand for computing, energy and data infrastructure grows.
The paradox of cheap AI judgment
The paradox extends beyond computing resources. If automated judgment becomes nearly free, organizations may hand AI many more decisions: every customer request, job application, transaction, insurance claim or employee message. Governments could use similar systems to prioritize applications, allocate services or flag cases for investigation.
This could raise productivity and widen access to services, particularly for small organizations and developing countries with limited specialist capacity. But it could also drive excessive automation. A structured answer is not necessarily a correct one, and a confidence score is not evidence or explanation. A wrong AI-written paragraph misleads a reader; a wrong AI judgment wired into software can reject an application, block a payment or remove content automatically. Even a small error rate causes real harm when repeated across millions of decisions.
Human oversight and accountability
The central question is not only whether Jev is accurate, but whether falling costs will push automated judgment into areas where decisions should remain explainable, contestable and subject to meaningful human review. Organizations will need to decide:
- which decisions AI may make on its own;
- which decisions require human approval;
- what level of uncertainty should trigger review;
- who is responsible when the system is wrong;
- whether affected people can challenge the result.
For developing countries, low-cost AI may broaden access to expertise and automation. It may also create dependence on systems built elsewhere that do not reflect local languages, laws or social conditions, and adoption may outpace local capacity for testing, regulation and appeal.
Purpose of the discussion
Using Jev as a case study, the session examines the wider consequences of cheaper, more efficient AI: whether efficiency will reduce resource use or stimulate demand; whether affordable AI will democratize access or accelerate over-automation; and what safeguards are needed when AI judgments directly affect people and institutions.
Questions for discussion
- Will cheaper AI reduce overall computing demand, or make AI use expand even faster?
- If AI judgment becomes almost free, will organizations automate decisions that do not need automating?
- Which decisions suit full automation, and which should always involve a human?
- Does a structured answer create a false impression that the judgment itself is reliable?
- Can a confidence score ever replace evidence, explanation and professional judgment?
- Should AI make consequential decisions if it cannot explain its reasons?
- Should people be told when AI influences decisions about employment, credit, insurance, education or public services?
- Should individuals have a right to challenge an automated decision and request human review?
- Who is responsible when an automated judgment causes harm?
- Will developing countries benefit from affordable AI, or become dependent on systems that do not reflect their realities?
Central question: will cheaper AI bring greater efficiency — or automate more decisions, consume more resources and multiply existing risks?
Organized by the AI for Developing Countries Forum (AIFOD), Geneva. Based on the AIFOD concept note for the session.
Cheaper AI won’t mean less AI — and human judgment is the real price
Avtandil Gagnidze, Professor and Vice-President of East-West University, joined an international panel at the AIFOD Weekly AI Review on the economics of AI, the Jevons paradox and the outsourcing of human judgment.
On 26 September 2026, the AI for Developing Countries Forum (AIFOD) held the 29th session of its Members’ Weekly AI Review online, under the title “Will Cheaper AI Automate Everything? The Economics of AI, the Jevons Paradox, and the Outsourcing of Human Judgment.” The panel’s shared conclusion was that falling AI prices are unlikely to shrink AI budgets, and may bring hidden costs: verification, vendor lock-in and the slow loss of the human expertise needed to check machine output.
The session was moderated by AIFOD Co-Founder and Board Director Tianze Zhang and brought together speakers from law, enterprise technology, education, global health and higher education. Opening the discussion, Zhang noted that the cost of GPT-3.5-level performance fell from roughly US$20 to about seven cents per million tokens between late 2022 and 2024, yet total spending depends on price multiplied by volume — the lesson of the Jevons paradox.
Prof. Avtandil Gagnidze: “AI cannot be responsible for a decision”
Professor Gagnidze argued that the key economic test is marginal benefit. “Price falls, but spending increases,” he said. “Because of diminishing returns, at some point we use more and the marginal benefit becomes zero or negative. Price is visible and easy to measure, but total cost includes invisible costs, like reputational risk, that most organizations never estimate.”
He stressed that bias at machine scale differs from human error: “Human mistakes are local, and we can recognize and correct them. An AI mistake can be repeated a trillion times.” Drawing on his work with medical doctors, he added:
“AI cannot be responsible for a decision. The decision-maker is always a concrete, identified person. The best approach is teamwork: let AI prepare decisions and offer alternatives, but the final decision must be made by a person.”Prof. Avtandil Gagnidze
He identified a structural gap as the main problem: the development of AI and the development of AI governance are moving at different speeds. In his closing message to developing countries, he said: “AI will not replace teachers, but teachers with AI skills will replace teachers without them. For small countries, digital poverty and language barriers are real problems, and we have to work with that reality.”
Other perspectives from the panel
Reda Sadki, Executive Director of The Geneva Learning Foundation, warned that cheaper generation means more to verify. He described a case in which his organization’s AI co-worker halted two reports after a verification script found quotations it could not match to the source data, and argued that automation stays safe only while people retain the skills to tell when the machine is wrong.
Fabrizio Degni, Chief AI Officer at Webuild, said token price is only one line in the true cost of AI, alongside infrastructure, people, data quality, governance, compliance and cybersecurity. “Benchmarks are a point of view, not the truth,” he noted, urging organizations to examine where a model comes from and how it was trained.
Dr Craig Gibbs, Executive Manager at JET Education Services, highlighted the risk of lock-in in the Global South, where a ministry may have only one chance to invest in an AI programme and cannot afford to switch even when it fails to meet local needs.
Valerie Pelton, AIFOD Senior Fellow and general counsel, argued that low prices must not mean lower standards, and that less advantaged communities risk getting what they can afford rather than what is best. “AI is a tool, not an end in itself,” she said.
About AIFOD
The AI for Developing Countries Forum is a Geneva-based global nonprofit working on AI governance, digital inclusion and AI sovereignty for the Global South, with more than 7,200 members in over 150 countries.
Read the full AIFOD report: Cheaper AI Won’t Mean Less AI, and Human Judgment Is the Real Price.
