Technology Forecasts 1998 vs 2026: The Fact Check

This fact check compares the 1999 book edition with the data status as of 17 July 2026.

The book covers surveillance, artificial intelligence, speech, handwriting and facial recognition, miniaturisation, evolutionary circuits, bio- and nanotechnology, telecommunications, bandwidth, ICANN, IPv6, e-commerce, cyborgs, wearables, augmented reality, body-area networks, acceleration, and the attention economy. We do not reproduce lengthy passages from the book or other historical texts here.

Table of Contents  

Results  

1. Artificial Intelligence  

Status in 1997/1998  

IBM Deep Blue won the match against the reigning world chess champion Garry Kasparov in 1997. The system used 32 processors, evaluated up to 200 million chess positions per second, and achieved 11.38 billion floating-point operations per second. It was highly specialised for chess.

For the book, this victory marked the beginning of a new phase of AI. It did not predict specific modern model architectures, generative AI, or a particular adoption rate.

Status in July 2026  

The Stanford AI Index 2026 cites several metrics for 2025: 88 per cent of surveyed organisations used AI, generative AI reached 53 per cent of the population within three years, and more than 90 per cent of the frontier models deemed significant originated from industry. On SWE-bench Verified, measured performance rose from 60 to nearly 100 per cent within a single year. At the same time, the best evaluated model read analogue clock times correctly in only 50.1 per cent of cases; agents on OSWorld achieved around 66 per cent of tasks, thus failing in roughly one out of three attempts.

Comparison  

The forecast of a new growth phase for AI is confirmed. The statements at the time were kept general; they did not contain a measurable prediction for generative models, multimodal systems, or an organisational adoption rate of 88 per cent. These subsequent developments are therefore not individually verifiable predictions, but rather a more robust expansion of the same technological category.

2. Speech, Handwriting, and Text  

Speech in 1999  

The book cites Dragon NaturallySpeaking and IBM ViaVoice as available systems. The recognition rate of around 95 per cent quoted there was a manufacturer claim after training, rather than an independent measurement comparable across products and languages. It lists pronunciation, speech tempo, slurred sounds, and ambient noise as unresolved challenges.

Speech in 2026  

Whisper was trained on 680,000 hours of multilingual and multitask supervised audio data. The model transcribes several languages and translates from multiple languages into English. The accompanying publication describes robust zero-shot performance but does not claim consistent error-free operation.

Handwriting and Documents in 1999  

The Palm Pilot with its proprietary input script serves as an example of that era. At the same time, the book notes that no general handwriting standard had yet prevailed.

Handwriting and Documents in 2026  

Google Cloud Vision offers handwriting recognition as a production OCR service. DOCUMENT_TEXT_DETECTION returns a hierarchy of page, block, paragraph, word, and character; asynchronous batch processing handles up to 2,000 image files. An EU endpoint is available for storage and processing, among other options.

Comparison  

Speech and text recognition are confirmed as computer interfaces. Only the format has changed: instead of a specifically learned input script on a single handheld device, today's services process free handwriting, scanned documents, and their structure. The old manufacturer claim of 95 per cent cannot be directly compared with today's model metrics, as datasets, languages, error rates, and testing conditions are missing.

3. Facial Recognition and Surveillance  

Facial Recognition from 1993 to 1998  

The US FERET programme ran from 1993 to 1998. The database contained 14,126 images of 1,199 individuals, captured in 15 sessions between August 1993 and July 1996. The programme received more than 6.5 million US dollars in funding; the development dataset was distributed to over 100 groups. Prior to FERET, recognition rates exceeding 95 per cent were frequently reported on controlled datasets containing fewer than 50 people. FERET introduced larger shared test datasets and independent evaluations.

The book already associated facial recognition with airports, security checkpoints, and automated surveillance.

Facial Recognition in 2026  

The ongoing NIST FRTE 1:1 evaluation recorded a total of 1,425 submitted algorithms from 437 developers by 15 June 2026. In 2026 alone, 62 algorithms from 61 developers had been added up to this reporting date. NIST measures the False Non-Match Rate at a fixed False Match Rate of 10⁻⁶. The evaluation continues to show variations depending on the dataset and image quality. NIST identifies insufficient lighting, over- or under-exposure, camera angles, and under-representation in the training data as causes of demographic differences.

Surveillance from 2024 to 2026  

The 2024 FTC report is based on information from nine major social media and streaming companies. The agency documents data collection on users and non-users, data from third-party sources and data brokers, indefinite retention in some cases, and the use of personal data in algorithms, analytics, and AI. Alongside visible camera surveillance, the investigation also covers clicks, advertisements, profiles, and recommendation systems.

The Digital Services Act sets the threshold for very large platforms at more than 45 million monthly active users in the EU. These platforms are required, among other things, to offer a non-personalised feed option. The DSA prohibits deceptive design patterns (dark patterns) and profile-based advertising targeted at children.

Comparison  

Facial recognition and automated surveillance are confirmed. The book focused on cameras, law enforcement, and identification. In 2026, commercial platform data, advertising profiles, and automated recommendation systems are demonstrably part of this landscape.

4. Miniaturisation  

Semiconductors in 1997/1998  

Intel introduced the Pentium II in 1997 with 7.5 million transistors, a 0.35-micrometre process, and clock speeds of 233, 266, and 300 MHz. 0.35 micrometres correspond to 350 nanometres. The book additionally mentions an IBM laboratory result of 1.1 GHz and structures of 0.18 micrometres.

Semiconductors in 2025/2026  

TSMC reported starting mass production of its N2 process in the fourth quarter of 2025. N2 utilises the company's first generation of nanosheet transistors.

The figures 350 nm and 2 nm do not represent geometrically identical measurements. For older processes, the number largely reflected an actual physical structure size; in contrast, modern node names bundle density, performance, and manufacturing characteristics, and do not correspond to any single gate length. While a factor of 175 could be calculated, it would be physically misleading.

Comparison  

The forecast of smaller and more powerful circuits is confirmed. What has not been confirmed is the idea that progress can be read solely from clock frequency: modern systems gain performance additionally through parallelism, specialised accelerators, memory architecture, and packaging.

5. Evolutionary Circuits  

Status in 1996/1998  

Adrian Thompson published an experiment in 1996 in which a genetic algorithm configured actual reconfigurable hardware. The book describes such evolutionary circuits as a possible alternative to circuits designed entirely by humans.

Status in 2026  

AlphaChip uses reinforcement learning to lay out chip blocks. According to Google DeepMind, the system generates layouts in hours rather than weeks or months, and was used in three generations of the Tensor Processing Unit. Optimisation occurs during the design phase, prior to manufacturing.

Comparison  

It is confirmed that learning or search-based methods are used in hardware design. What is not confirmed is hardware that evolutionarily rebuilds its physischal circuitry during runtime. The verdict is therefore partially confirmed.

6. Bio- and Nanotechnology  

DNA Computing in 1994/1998  

Leonard Adleman solved a combinatorial path problem using DNA molecules in 1994. The experiment demonstrated that molecular reactions could be used for computation. From this, the book extrapolated far-reaching possibilities, including personal biochips and a potential replacement of conventional computers.

Biocomputing in 2025/2026  

A 2025 review in Nature Reviews Bioengineering describes DNA-based logic gates and circuits for biosensing, diagnostics, and conditional therapeutic responses. The academic literature identifies robustness, scalability, standardisation, and clinical translation as unresolved challenges. A general DNA computer acting as a replacement for PCs, smartphones, or servers is not documented.

The FDA approved Casgevy on 8 December 2023 as the first authorised therapy using CRISPR/Cas9. Of 31 evaluable individuals with sickle cell disease, 29 (representing 93.5 per cent) achieved at least twelve consecutive months free of severe vaso-occlusive crises. The treatment modifies blood stem cells and is a gene therapy, not a universal biocomputer.

Nanotechnology in 2026  

The National Nanotechnology Initiative documents applications in transistors and memory, displays, sensors, catalysts, water purification, drug delivery, diagnostics, coatings, batteries, and lightweight materials. These applications do not form a single product segment or a uniform computing architecture.

Comparison  

The prediction of a general biological computer replacement is unconfirmed. DNA-based computation and genome editing are confirmed as specialised laboratory and medical procedures. The nanotechnology forecast is partially confirmed, as applications are demonstrable in several industries, but the book did not define a measurable target value or timeline.

7. Bandwidth and Telecommunications  

Status in 1998  

The ITU approved V.90 on 1 September 1998. The modem standard specified up to 56,000 bit/s from the digital network to the analogue subscriber line and up to 33,600 bit/s in the opposite direction.

The book also mentions an optical laboratory experiment using 3.8-femtosecond pulses, deriving a theoretical rate of 263 Gbit/s. It predicts that data services will become the core business of telecommunications and cites a contemporary forecast of a 90 per cent data share of mobile traffic in 2005.

Status from 2024 to 2026  

IEEE 802.3df-2024 standardises Ethernet at 400 and 800 Gbit/s. Mathematically, 800 Gbit/s is roughly 14.3 million times faster than 56 kbit/s. However, this factor contrasts a modem access standard with a modern high-performance network Ethernet standard and is therefore not a comparison of end-user speeds. Compared to the theoretical optical rate of 263 Gbit/s mentioned in the book, the standardised 800 Gbit/s value is about 3.04 times higher.

Ericsson dates the global intersection of mobile data and voice volume to late 2009: only then did monthly data volume exceed voice traffic for the first time. The forecast of 90 per cent data traffic by 2005 cited in the book was therefore incorrect in both timing and scale; only the subsequent direction of development was confirmed.

The ITU reports almost three-quarters of the global population online in 2025, with 2.2 billion people offline. 5G covered more than half of the world's population and accounted for more than a third of mobile broadband connections. The ITU continues to highlight gaps in quality and affordability.

Comparison  

The rapid increase in bandwidth is confirmed. The specific prediction of 90 per cent data traffic by 2005 is unconfirmed. The fact that mobile data volume exceeded voice traffic in late 2009 confirms the direction of development, albeit with a delay of at least four years relative to the targeted year. The verdict for 'data over voice' is therefore partially confirmed.

8. ICANN and Internet Governance  

Status in 1999  

ICANN's articles of incorporation were filed on 30 September 1998. Looking back, ICANN estimated the number of internet users at that time to be approximately 100 million. The book also takes into account the first fiscal year, Network Solutions, and the Shared Registry System.

Status in 2016/2026  

In October 2016, the contract between the US government and ICANN regarding the IANA functions expired. Stewardship transitioned to the global multistakeholder model. ICANN continues to coordinate the Domain Name System and internet numbers; state regulation of data protection, platforms, competition, and content occurs in parallel at national and regional levels.

The ITU figure of almost three-quarters of the global population online equates to roughly six billion people. Compared to the roughly 100 million in 1998, this is an increase of approximately 60-fold; data collection methods and definitions have changed over this period.

Comparison  

The prediction that internet governance would become a permanent global coordination and policy task is confirmed. The book did not predict sole authority for ICANN; in 2026, technical multistakeholder coordination and territorial law enforcement coexist.

9. IPv6  

Status in 1998  

RFC 2460 was published in December 1998. IPv6 expanded the address length from 32 to 128 bits and defined, among other things, a simplified header, extension headers, and flow labels. RFC 2460 was superseded by RFC 8200 in 2017. The book linked the larger address space to a growing number of networked devices and wearable computers.

Status in July 2026  

Google measures the proportion of its own users who access Google via IPv6. For 15 July 2026, the published dataset shows 47.287251 per cent; on 11 July, the figure was 50.246537 per cent. The fluctuation is due, among other things, to the difference between weekdays and weekends. This metric does not measure total global traffic, but rather Google usage alone.

Comparison  

The need for a larger address space and the active use of IPv6 are confirmed. A complete replacement of IPv4 has not occurred. The verdict is partially confirmed because the technical rationale is correct, yet coexistence persists at least 28 years after the publication of the specification.

10. E-Commerce  

Start of Measurement in 1999  

The U.S. Census Bureau began monthly data collection on retail e-commerce in October 1999. The first official quarterly estimate was published on 2 March 2000: in the fourth quarter of 1999, e-commerce accounted for 5.3 billion US dollars, or 0.6 per cent of US retail sales. A methodologically identical official figure for the fourth quarter of 1998 is not available in this series.

The book describes e-commerce as an end-to-end process comprising information, catalogue, ordering, payment, delivery, and customer service.

Status in 2025/2026  

For the first quarter of 2026, the U.S. Census Bureau reports seasonally adjusted, non-price-adjusted e-commerce retail sales of 326.7 billion US dollars. This represented 16.9 per cent of total retail sales of 1,929.0 billion US dollars. Compared to the first quarter of 2025, e-commerce increased by 9.8 per cent, while total retail sales grew by 3.9 per cent.

The sales comparison of 5.3 to 326.7 billion US dollars is not a real growth factor due to inflation, seasonal adjustments, quarterly differences, and methodological changes. The share of retail sales rose from 0.6 to 16.9 per cent; this represents an increase of 16.3 percentage points, or approximately 28.2 times the baseline share.

For the EU, Eurostat reports for 2025: 95 per cent of 16 to 74-year-olds had used the internet in the preceding twelve months; 78 per cent of these internet users had ordered goods or services online. 63 per cent of online shoppers reported experiencing no problems during their purchases.

Comparison  

The forecast of an end-to-end digital commerce process is confirmed. Available statistics substantiate both a higher share of sales and widespread use in the EU.

11. Cyborgs, Wearables, and Body-Area Networks  

Body-Area Communication in 1996/1998  

In 1996, Thomas Zimmerman at IBM described a Personal Area Network using near-field capacitive communication through the human body. A prototype transferred electronic business cards via a handshake. The book linked this research to wearables, smart clothing, jewellery, military systems, and a closer integration of body and computer.

Wearables in 2026  

The FDA lists over-the-counter ECG software for handheld or wrist-, arm-, or chest-worn electronics as a Class II medical device. The software is permitted to generate, analyse, and display ECG data and provide notifications of arrhythmias; it is explicitly not intended for independent diagnosis.

This class of devices demonstrates that wearable sensors and regulated software are now firmly integrated. It does not show that smart clothing or internal body data networks have become a general standard.

Brain-Computer Interface in 2024  

In an NIH-funded study, four arrays of 64 microelectrodes each, totalling 256 electrodes, were implanted into the left precentral gyrus of a person with ALS. After 30 minutes of calibration, the system recognised more than 99 per cent of words from a 50-word vocabulary. With a 125,000-word vocabulary, recognition exceeded 90 per cent after further calibration. After 16 hours of use, the system reached approximately 97.5 per cent accuracy and maintained this level for over eight months; the speech rate was about 32 words per minute.

Comparison  

Wearable sensors, medical implants, and direct neural interfaces are confirmed. What has not been confirmed is the mainstream 'cyborg' as a widespread physical enhancement for healthy individuals. Smart clothing and body-area data transmission remain niche forms compared to watches, rings, headphones, and medical devices. Cyborgs, wearables, and body-area networks are therefore each evaluated as partially confirmed.

12. Augmented Reality  

Status in 1997/1998  

Ronald Azuma's 1997 overview defined Augmented Reality by combining real and virtual content, real-time interaction, and three-dimensional registration. Applications in medicine, manufacturing, visualisation, entertainment, and the military were noted. Technical challenges included tracking and registration errors. The book adopted AR as a theme for context-aware information and body-worn computers.

Status in 2024/2026  

Apple launched the Vision Pro in the US on 2 February 2024. The 256 GB model started at 3,499 US dollars, with optical inserts costing an additional 99 or 149 US dollars. Interaction combines eyes, hands, and voice.

The market launch demonstrates that a spatial computer is technically and commercially available. Price and availability say nothing about widespread everyday adoption. We have not found a reliable, cross-manufacturer adoption rate for AR headsets for this article and therefore make no such claim.

Comparison  

The technical execution of the AR forecast is confirmed. The anticipated mainstream everyday role of wearable AR systems is not substantiated by the data used. The verdict is partially confirmed.

13. Attention Economy  

Status in 1997/1998  

Michael Goldhaber published "The attention economy and the Net" on 7 April 1997 in First Monday. The article treated attention as the central scarce resource of the net. The book incorporated the term into its technology outlook.

Status in 2026  

On 6 February 2026, the European Commission preliminarily found that TikTok's addictive design might violate the Digital Services Act. The investigation identifies four specific design features: infinite scroll, autoplay, push notifications, and a highly personalised recommendation system. The finding is preliminary and does not prejudge the final outcome of the proceedings.

The DSA requires very large platforms with more than 45 million monthly active users in the EU to provide a non-personalised feed option and prohibits dark patterns. Thus, the selection, ordering, and presentation of content are explicitly subject to statutory obligations.

Comparison  

The forecast that attention on the web is managed economically is confirmed. The book did not specify a metric for screen time, advertising revenue, or recommendation share. The confirmation therefore concerns the business and design principles, rather than a quantitative prediction.

14. Acceleration  

The book addresses shorter innovation cycles, faster communication, and increasing time pressure under the concept of "acceleration". It defines neither an index, a target value, nor a timeframe for this.

Depending on whether one uses processor performance, bandwidth, volume of messages, working hours, productivity, or subjective time pressure, a different statement is evaluated. We therefore assess this topic as not quantitatively verifiable. Individual metrics from other chapters—800 Gbit/s Ethernet, 88 per cent AI usage in organisations, or 326.7 billion US dollars in quarterly e-commerce sales—do not replace a missing definition of "acceleration".

Overall Balance  

The balance evaluates topics, rather than individual sentences from the book. Broad statements do not receive a stronger verdict than the available metrics support. We identify manufacturer specifications as such. Non-comparable metrics—such as 56 kbit/s modem access and 800 Gbit/s Ethernet—are not represented as identical end-user measurements.

Historical retrospective on the 1999 book