Public-Record Analysis | Technology, Science & Public Accountability
Artificial Intelligence Is 70 Years Old. Could Its Next Advances Come Much Faster?
Researchers were already considering machines that could improve themselves in 1955. Today, AI helps develop computing algorithms, advanced models are improving rapidly, and the infrastructure supporting them is expanding. The next question is whether those developments will reinforce one another — and what they will mean for Texans.
Artificial intelligence is beginning to help develop the technology that will support future artificial intelligence.
The idea is not new.
On August 31, 1955, four researchers proposed a summer project to investigate whether machines could perform activities associated with human intelligence. Among the questions they wanted to explore was whether machines could improve themselves.
In 2026, 70 years after the resulting Dartmouth workshop, researchers are demonstrating systems that help discover and test improvements to computing algorithms.
At the same time, the computing resources supporting AI are expanding rapidly.
Stanford University's 2026 AI Index estimates that global AI computing capacity grew approximately 3.3-fold annually from 2022 through 2025.
The International Energy Agency estimates that worldwide data-center electricity consumption reached 485 terawatt-hours in 2025. Its central projection places consumption at 950 terawatt-hours by 2030.
Those figures describe different aspects of technological development. Neither establishes that intelligence itself is increasing at a fixed rate.
Together, however, they raise a consequential question:
Could increasingly capable AI systems help accelerate the development of their successors?
The answer matters not only to researchers and technology companies.
For Texans, the expansion of AI is already connected to electricity infrastructure, government technology oversight, public spending and questions about how much authority should be delegated to automated systems.
A 70-year-old idea
Artificial intelligence did not begin with ChatGPT.
Its intellectual foundations extend back decades before the internet became part of everyday life.
In 1950, British mathematician Alan Turing published Computing Machinery and Intelligence in the journal Mind.
The paper examined whether machines could demonstrate intelligent behavior and proposed evaluating their performance through conversation.
Five years later, John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon submitted a proposal for the Dartmouth Summer Research Project on Artificial Intelligence.
Their objective was to investigate whether aspects of intelligence could be described precisely enough for machines to reproduce them.
They proposed studying language, problem-solving, concept formation and machine self-improvement.
The researchers gathered at Dartmouth College in New Hampshire during the summer of 1956.
That workshop is widely recognized as the formal beginning of artificial intelligence as a scientific discipline. The 70-year milestone refers to 1956; the proposal was written the year before.
The computers available at the time were limited compared with modern systems.
Yet the original research questions remain recognizable.
Could machines learn?
Could they reason?
Could they solve unfamiliar problems?
And could they eventually help improve their own capabilities?
The last question is particularly relevant today.
What changed over seven decades?
AI did not advance in a straight line.
Early systems concentrated on symbolic reasoning and explicitly programmed rules.
Later approaches included expert systems designed to reproduce specialized human decision-making.
Machine learning expanded the ability of computers to identify patterns in information rather than relying exclusively on instructions written for every possible situation.
Advances in neural networks, specialized processors and deep learning eventually supported major improvements in language processing, image recognition and other applications.
By the 2010s, AI had become embedded in services used by millions of people.
Search engines, navigation applications, automated translation and recommendation systems increasingly relied on machine-learning techniques.
Many people were using artificial intelligence without recognizing it.
The public expansion of generative AI during the 2020s made the technology much more visible.
People could directly interact with systems capable of producing text, images, computer code and other content.
AI had not suddenly appeared.
Its capabilities had become accessible in a different way.
What makes a system artificial intelligence?
Not every automated computer system is AI.
A conventional program may follow a predetermined set of instructions.
An AI system may instead identify patterns in data, make predictions, generate content or select actions based on learned relationships.
Consider email filtering.
A simple automated rule might move every message containing a particular phrase into a designated folder.
A machine-learning system can analyze patterns across messages and estimate which are likely to be unwanted, including messages containing unfamiliar wording.
The distinction is useful because sophisticated automation existed long before today's AI systems.
Automation and artificial intelligence can overlap, but they are not interchangeable.
AI is already performing complex work
Artificial intelligence is no longer confined to specialized laboratories.
AI systems assist with scientific research, software development, medical-image analysis, education and industrial operations.
Some can use software tools, evaluate intermediate results and complete sequences of tasks.
Stanford's 2026 AI Index documents substantial improvements across selected scientific, mathematical, reasoning and coding benchmarks.
But those improvements are uneven.
The report describes a striking example: AI systems capable of exceptional mathematical performance can still struggle with tasks such as reading analog clocks.
This illustrates what researchers sometimes call AI's jagged frontier.
A system may perform extremely well on one category of problems while remaining unreliable on another.
Strong benchmark performance does not establish that AI possesses every aspect of human intelligence.
Nor does it establish that the system can operate safely without supervision.
AI can produce convincing explanations containing factual errors. It can misunderstand instructions or overlook information that changes the correct answer.
The practical question is therefore not simply whether AI can perform a task.
It is whether the system can perform that task reliably under the conditions in which people intend to use it.
That distinction becomes especially important when AI influences decisions involving public services, finances, safety or individual rights.
The computing infrastructure behind AI
Modern AI depends on substantial physical infrastructure.
Data centers contain computing equipment, networking systems and specialized processors used to train and operate AI models.
They also require electricity and cooling.
As developers pursue more capable systems, they are investing in additional computing capacity.
Stanford's 2026 AI Index estimates that global AI computing capacity grew roughly 3.3 times per year from 2022 through 2025.
The underlying measure concerns AI chips supplied by major designers, expressed in equivalent computing capacity. The report cautions that the estimates are inferred from revenue data, financial disclosures and analyst reports, with coverage varying by manufacturer.
That is an extraordinary rate of growth in estimated computing resources.
It is not a measurement showing that intelligence has increased by the same factor, or that every individual model's computing resources grew at that rate.
A larger computing system can perform more calculations and support more demanding workloads.
But effective AI also depends on algorithms, training methods, data quality and system design.
A useful comparison is a factory.
Adding more machinery can increase production capacity.
But production also depends on how efficiently the machinery operates and how well the manufacturing process is organized.
Data centers provide computational capacity. They do not automatically convert every additional processor into an equivalent improvement in intelligence.
The distinction matters when evaluating claims about how quickly AI might advance.
Does AI learn everything it encounters?
The expansion of computing capacity is often discussed alongside the growing amount of information available to AI systems.
Large datasets can help models identify patterns across language, images, scientific information and other material.
But more information does not necessarily produce better learning.
Repeated, inaccurate or irrelevant information may contribute little useful knowledge.
There is also an important difference between using information and permanently learning from it.
Many deployed AI models operate using parameters established during training.
When someone provides information during a conversation, the system may use that information to generate a response.
That does not necessarily mean the information permanently changes the underlying model.
Some systems use memory or retrieve information from external sources.
Those capabilities can improve their usefulness without changing the model's fundamental training.
Developing more capable AI may require additional training, improved algorithms or modifications to the system itself.
That is why expanding data centers and providing more information are important developments, but not sufficient explanations for future AI progress.
AI is beginning to help improve computing technology
In May 2025, Google DeepMind introduced AlphaEvolve.
The company described it as an AI-assisted system designed to propose improvements to algorithms and evaluate those proposals through automated testing.
Its reported applications include mathematics and computing optimization.
The significance goes beyond generating computer code.
A system that proposes alternative algorithms can test them against measurable criteria.
If an alternative performs better, researchers may incorporate that improvement into operational technology.
For example, an algorithm that completes a calculation using fewer computing resources could make existing hardware more efficient.
That improvement could then support additional research.
Google DeepMind's reported results come from the technology's developer. They should not be interpreted as independent proof of unlimited machine self-improvement.
Nevertheless, AlphaEvolve provides a documented example of AI contributing to technical development.
Three different processes must be distinguished.
AI can assist human researchers.
Automated systems can propose and evaluate improvements within defined tasks.
A system independently redesigning and deploying more capable successors without human supervision would represent a substantially different level of autonomy.
The available AlphaEvolve evidence supports the first two processes.
It does not establish the third.
Could AI progress compound?
Compounding occurs when improvements build upon earlier improvements.
Compound interest offers a familiar example.
An investment earns interest. When that interest remains invested, subsequent gains are calculated on a larger balance.
Technological development can sometimes follow a related pattern.
A successful innovation creates tools that make additional innovations possible.
AI could strengthen that process.
Additional computing capacity allows researchers to conduct more experiments.
Improved algorithms can make computing resources more efficient.
More capable AI systems can assist with programming, analysis and technical design.
Successful improvements may then support later generations of AI.
Imagine researchers developing an AI system that helps programmers complete certain tasks more efficiently.
Those programmers use the assistance to improve research software.
Better research software allows more experiments to be evaluated.
Some experiments identify more efficient algorithms.
Those algorithms may then support the development of more capable AI systems.
Each successful improvement contributes to the conditions supporting the next.
That is a plausible technological feedback loop.
But it is not guaranteed.
An AI system may generate thousands of proposals without identifying a useful improvement.
Scientific discoveries may require physical experiments, specialized equipment or knowledge that cannot be obtained through computation alone.
Progress can also encounter diminishing returns.
AI-assisted development creates a mechanism through which technological progress could accelerate. It does not establish that acceleration will continue indefinitely.
The decisive evidence would be sustained, independently measured improvements in research productivity, development time, system reliability and the performance of successive generations.
The electricity required for expansion
The International Energy Agency's April 2026 report, Key Questions on Energy and AI, documents the scale of data-center expansion.
The agency estimates that global data-center electricity consumption increased 17% in 2025.
Electricity consumption by AI-focused data centers increased approximately 50% that year.
The IEA estimates that all data centers worldwide consumed approximately 485 terawatt-hours in 2025.
Its central projection places that figure at approximately 950 terawatt-hours in 2030.
That represents an increase of about 96%.
A terawatt-hour is one billion kilowatt-hours.
For comparison, a household using 1,000 kilowatt-hours per month consumes 12,000 kilowatt-hours annually.
One terawatt-hour is equivalent to the annual electricity consumption of approximately 83,000 households using that amount.
The comparison illustrates scale. It does not establish how data-center electricity costs are distributed among customers.
The IEA also warns that infrastructure bottlenecks could constrain development.
Electricity availability, grid connections, specialized equipment, chip manufacturing and financing all influence how quickly data centers can expand.
The agency's projections are not guarantees.
They also do not establish the electricity consumption or financial consequences of any particular Texas project.
Those questions require project-specific evidence.
Is military AI ahead of what the public can use?
Government and military organizations have supported artificial intelligence research for decades.
Some capabilities are classified.
Others are developed through publicly documented programs or partnerships with commercial companies.
But advanced automation and artificial intelligence should not be confused.
A system that automatically calculates missile steering corrections is not necessarily using AI.
Military technology also moves through development, testing, procurement and operational deployment.
A weapons system may enter service years before it reaches every unit.
Older equipment may remain in use while newer technology is being introduced.
That history demonstrates a difference between technological development and operational availability.
It does not establish that classified AI is universally decades ahead of commercial systems.
The Defense Department's own procurement record complicates that assumption.
On July 14, 2025, its Chief Digital and Artificial Intelligence Office announced awards involving four commercial frontier-AI developers: Anthropic, Google, OpenAI and xAI.
The announcement described a contract ceiling of $200 million for each award. Those ceilings represent maximum potential contract values, not proof that the full amounts were spent.
The department said the awards would accelerate military adoption of advanced AI capabilities.
The announcement demonstrates that military organizations are seeking technology developed by commercial companies.
Some government capabilities may remain unavailable to the public.
Other technologies may be commercially available before military organizations complete their own evaluation and procurement processes.
The public records examined for this article do not establish that classified military AI is universally 10, 20 or 30 years ahead of commercially available AI.
The comparison depends on the specific capability being examined.
Does increasing intelligence mean increasing awareness?
As AI systems become more capable, questions about awareness and consciousness become increasingly important.
But the terms describe different concepts.
A system can monitor information, recognize changes, evaluate aspects of its performance and revise its responses.
These functions may be described as self-monitoring or functional awareness.
Subjective consciousness is a different question.
It concerns whether a system actually experiences anything from its own perspective.
A thermostat detects temperature and responds when conditions change.
That does not establish that it experiences warmth or cold.
An AI system can perform far more sophisticated monitoring.
It may describe its capabilities, identify inconsistencies and revise previous answers.
Those behaviors demonstrate information processing.
They do not independently establish subjective experience.
A machine might accurately explain an emotion without experiencing it.
It might discuss its own existence without possessing an internal sense of being alive.
Researchers are investigating whether artificial systems could possess consciousness and how that possibility might be evaluated.
In a paper appearing in the June 2026 issue of Trends in Cognitive Sciences, Patrick Butlin and fellow researchers examined indicators derived from scientific theories of consciousness. The paper was first published online on November 10, 2025.
The researchers emphasize significant uncertainty about relevant issues in consciousness science.
Their work proposes a method for evaluating relevant indicators while acknowledging that uncertainty.
It does not establish that current AI systems are conscious.
Nor does it prove that machine consciousness is impossible.
The scientific distinction remains essential.
AI can become more capable, autonomous and sophisticated without researchers establishing that it possesses subjective consciousness.
What could the next 70 years bring?
Artificial intelligence's history demonstrates substantial technological change.
It does not provide a reliable timetable for future breakthroughs.
Many developments once imagined as distant possibilities already exist in specialized or limited forms.
AI assists scientific research.
It contributes to software development.
It supports industrial automation.
It helps researchers evaluate complex technical problems.
The next stage may involve making those capabilities more reliable, more autonomous and more widely available.
AI-assisted research could shorten certain development cycles.
More efficient algorithms could make existing computing resources more productive.
Advances in hardware could expand the experiments researchers can conduct.
But future development may encounter unexpected limits.
Energy requirements, computing costs, reliability problems and the difficulty of translating research results into practical applications could slow progress.
Some breakthroughs may occur sooner than expected.
Others may prove much harder to achieve.
The next 70 years could produce AI systems dramatically different from today's technology.
What cannot be established is exactly how quickly those changes will occur or whether they will include subjective machine consciousness.
Texas has already begun regulating AI
For Texans, artificial intelligence is not merely a discussion about future technology.
It is already part of state law and government reporting.
House Bill 149, enacted by the 89th Texas Legislature, took effect January 1, 2026.
Among its provisions, the legislation amended Texas Government Code §2054.0965 to include artificial intelligence systems in the inventory of technology applications maintained through the state's Information Resources Deployment Review.
The Texas Department of Information Resources' 2026 reporting instructions specifically address AI inventories and heightened-scrutiny AI systems.
The agency identified March 31, 2026, as the deadline for state agencies to submit their reviews.
Senate Bill 1964, effective September 1, 2025, established additional requirements concerning public-sector AI governance and risk management.
In an August 2026 implementation update, DIR reported adopting rules in Title 1, Texas Administrative Code, Chapter 219, and developing AI governance resources.
Those are documented government actions.
They do not establish how many AI systems each state agency currently operates, whether every system has been properly reported or how effectively individual deployments are supervised.
Those questions require examination of the underlying inventories, agency records and implementation results.
The distinction matters because requiring agencies to identify AI systems is not the same as demonstrating that every use of AI is safe, effective or accountable.
Who pays for AI's physical infrastructure?
Texas also faces questions involving data-center development.
Large computing facilities may require substantial electricity capacity, cooling infrastructure and utility connections.
Depending on the project, additional transmission or distribution equipment may be needed.
But the existence of a data center does not automatically establish that residential customers will pay for those upgrades.
Some infrastructure costs may be assigned directly to developers.
Others may be recovered through utility rates under applicable regulatory arrangements.
Determining the financial consequences requires examining the specific project, relevant utility filings and approved cost-allocation decisions.
The same principle applies to economic incentives.
An announced investment is not completed construction.
Projected employment is not verified job creation.
An approved tax incentive is not proof that the promised public benefit has been delivered.
For Texans, the appropriate questions are concrete.
What infrastructure is required?
Who pays for it?
What benefits have actually materialized?
And what do the controlling public records establish?
Those questions can be investigated regardless of whether future AI systems become dramatically more capable.
The newsroom reporting this story relies on AI
There is an irony in this investigation.
The Texas Public Record relies heavily on artificial intelligence to help produce journalism examining artificial intelligence itself.
The publication uses AI to assist with research, public-record analysis, identifying discrepancies, article development and newsroom operations.
That reliance illustrates how the technology is changing what small organizations can accomplish.
It also demonstrates why human oversight remains necessary.
During development of this article, an AI-generated projection initially placed several existing technological capabilities decades into the future.
Human editorial review identified the problem and prompted a correction.
The incident illustrates both the value and limitations of AI-assisted journalism.
AI can organize complex information, help analyze records and identify questions worth investigating.
It can also make incorrect assumptions, misinterpret evidence or express uncertain conclusions with excessive confidence.
Using AI does not transfer responsibility for accuracy to the technology.
The newsroom must still determine whether material claims are supported, whether records have been interpreted correctly and whether necessary qualifications have been included.
For The Texas Public Record, publication decisions remain the responsibility of its human editor.
This article was developed with substantial AI assistance.
The disclosure matters because TPR is not simply observing AI's development.
It is participating in it.
The broader implication extends beyond journalism.
Increasingly accessible AI may expand what individuals and small organizations can accomplish without requiring the resources of major corporations.
Whether those gains are reliable and beneficial depends on how the technology is used, evaluated and supervised.
The next chapter is already underway
In 1955, researchers proposed investigating whether machines could improve themselves.
Seven decades later, AI systems are contributing to algorithm development while computing infrastructure expands rapidly.
Those developments do not establish that technological progress will accelerate indefinitely.
They do demonstrate that AI is becoming a participant in some of the research and engineering processes used to develop future technology.
The question is no longer whether artificial intelligence can contribute to technological advancement.
It already does.
The question is how far those contributions can extend, how reliably improvements can be demonstrated and whether future systems will make subsequent development substantially faster.
For Texans, the immediate responsibilities are more concrete.
Public officials must account for government AI deployments.
Infrastructure decisions must be supported by accurate cost and benefit information.
And organizations using AI must remain responsible for the consequences of that use.
Bottom line
Artificial intelligence has evolved substantially since 1956. AI-assisted research and expanding computing resources could accelerate future advances, but unlimited technological progress and subjective machine consciousness remain unproven. Texans can assess the technology's public consequences through government records, infrastructure decisions, documented costs and accountable oversight.
Sources
Primary records
- Alan Turing — Computing Machinery and Intelligence, Mind, October 1950
- King’s College, Cambridge — Turing Digital Archive, AMT/B/9, original paper catalogue and provenance
- McCarthy, Minsky, Rochester and Shannon — A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, August 31, 1955; John McCarthy’s Stanford archive
- Dartmouth College — The Founding and Future of AI; 1956 workshop
- Texas Legislature — HB 149, 89th Regular Session, enrolled text, §§7 and 10; Government Code §2054.0965 inventory amendment and January 1, 2026 effective date
- Texas Legislature — HB 149 legislative history and enacted status
- Texas Legislature — SB 1964, 89th Regular Session, enrolled text; Government Code Chapter 2054 AI governance provisions and September 1, 2025 effective date
- Texas Legislature — SB 1964 enrolled bill summary and effective date
- Texas DIR — 2026 Information Resources Deployment Review Instructions, January update; AI inventory guidance, printed pp. 63–64
- Texas DIR — 2026 IRDR and IRCAP; March 31 reporting deadline and AI inventory instructions
- Texas DIR — AI in Texas: DIR Implementation of Laws from the 89th Legislature, August 14, 2026
- Texas Secretary of State — Title 1, Part 10, Chapter 219, Artificial Intelligence; official rule locator linked by DIR
- U.S. Defense Department CDAO — Frontier-AI award announcement, July 14, 2025; indexed announcement text, direct retrieval unavailable during this check
- U.S. Senate — September 10, 2025 correspondence identifying the four CDAO award recipients and citing the July 14 announcement; used only to corroborate award history
Research and analysis
- Stanford University — 2026 AI Index Report; technical performance, science, medicine and education
- Stanford University — 2026 AI Index, Chapter 1, Research and Development, printed pp. 29–30; estimated global computing capacity and methodology
- International Energy Agency — Key Questions on Energy and AI, April 16, 2026
- International Energy Agency — Executive summary; 2025 estimates, central 2030 projection and infrastructure constraints
- Butlin and coauthors — Identifying Indicators of Consciousness in AI Systems, Trends in Cognitive Sciences, June 2026 issue; first published online November 10, 2025
- PubMed — Butlin paper bibliographic record and author abstract; theory-derived indicators and uncertainty
- Lewis and coauthors — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, NeurIPS 2020; parametric knowledge and external retrieval
Interested-party statements
- Google DeepMind — AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms, May 14, 2025
- OpenAI — Introducing OpenAI for Government, June 16, 2025; CDAO contract and $200 million ceiling
Data and methodology
Stanford’s 3.3× figure concerns estimated global AI computing capacity across major chip designers during 2022–2025. It is not a direct measure of intelligence or the training resources used by every individual model. Manufacturer coverage varies, and the estimates are inferred from revenue data, financial disclosures and analyst reports.
The International Energy Agency’s 2025 electricity figure is an estimate; its 2030 figure is a projection. The projected increase is calculated as (950 − 485) ÷ 485 × 100, or approximately 95.9%.
One terawatt-hour equals one billion kilowatt-hours. At an illustrative consumption level of 12,000 kilowatt-hours per household annually, that equals approximately 83,333 household-years of electricity use. The comparison does not measure actual Texas household consumption or establish electricity cost allocation.
Source-access note: The Dartmouth proposal was inspected in John McCarthy’s Stanford archive. Turing’s original publisher page was not directly retrievable; its publication details were corroborated through the King’s College archive catalogue. The consciousness paper’s bibliographic record and author abstract were inspected through PubMed. The CDAO award announcement was available as indexed text, but direct retrieval was unavailable; the award history was corroborated against official Senate correspondence and OpenAI’s own contract announcement. The current rule portal returned no readable text, so the Chapter 219 adoption statement is attributed to DIR’s August 14 update rather than presented as a separate audit of the rules. The newsroom-development incident is the editor’s account supplied with this article.
Verified through October 30, 2026.
EDITORIAL NOTICE: This report reflects records and information available at publication. Allegations, charges, and disputed claims are not findings of wrongdoing. Submit documented correction or clarification requests to The Texas Public Record. Material errors will be corrected transparently.
The Texas Public Record
Rights. Accountability. Awareness.