Agencies
As the latest wave of artificial intelligence systems rapidly comes under the scrutiny of legal and policy frameworks, a captivating question is taking center stage: How can AI forget what it has learned? Can the European Union’s "right to be forgotten” regulations also apply to AI? In the background, a classic Turkish art music song by Nesrin Sipahi plays on a record: "That first glance that fills the heart is unforgettable, unforgettable.”
Indeed, every song, poem and work of art is as much a collective memory of society as it is of the artist. But should AI systems also share this memory? Already plagued by copyright issues, AI is now facing challenges in the domain of learning as well. Previously, the concept of the "right to be forgotten” in search engines and social media, as defined by the European Union, sparked debates on whether it could be extended to AI.
The rule allows individuals to request that organizations remove and delete personal information about them from online platforms.
An article by Politico discusses this issue in-depth.
Given the immense energy required for AI systems to learn faster and more efficiently while ingesting vast amounts of information, expecting them to forget may seem like an odd goal.
However, the owners of this information also have a stake in the matter. As the number of publishers signing licensing agreements with ChatGPT-maker OpenAI increases, questions arise about what will happen when these agreements expire, how publishing houses will retract access and whether it is technically possible to erase their editorial content from future queries.
Regulators in Europe aim to apply existing digital law, the Digital Services Act, to new platforms. This might mean that AI would need to "forget” information it has memorized about people.
Difficult but not impossible So, is there a way to erase what an AI system has already learned without retraining the model from scratch? This is a much more complex issue than deleting pieces of information from a database.
In fact, a new research area called machine learning has emerged in recent years to find out what methods can be used to make AI models selectively and retrospectively forget or at least approximate their training data.
Researchers claim that if they can develop easier ways for AI to completely forget information, it would facilitate the clean implementation of these agreements and regulations.
It’s not just the new generative AI models that have sparked this question.
Researchers began contemplating machine unlearning after the EU’s 2014 ruling granting residents the legal "right to be forgotten,” allowing them to request the deletion of personal data from internet searches and other digital records.
Generative AI models teach themselves from existing data patterns using neural networks that mimic the human brain. Even the scientists who create these models can’t fully explain how this self-learning happens, making it nearly impossible to create algorithms that can reverse the learning process.
They need to prevent the influence of specific data while resisting attempts to extract this information through jailbreak attacks and maintaining the model’s overall performance.
One of Türkiye’s leading telecommunications and technology companies, Türk Telekom, has said its consolidated revenues rose by 6% annually in the first quarter of the year to reach TL 28 billion ($867.18 million).
The financial results showed its EBITDAearnings before interest, taxes, depreciation and amortization – increased by 23% to TL 10.3 billion, while its EBITDA margin improved by 510 basis points to 36.6% compared to the previous year.The company ended 2023 as the investment leader in its sector. From January through March, it injected another TL 4.6 billion, and its financial expenses reached TL 6.9
million.
The number of households using fiber in Türkiye continues to increase, as highlighted by Türk Telekom’s results.
According to the company, the number of its mobile subscribers reached 26.2 million and the number of households with fiber access extended to 32.2 million.
Türk Telekom continues to expand its fiber infrastructure across all 81 provinces in Türkiye, preparing the country for future technologies.
As of the first quarter, its fiber network had reached 441,000 kilometers (274,024 miles)."With our technological expertise and investments, we are leading Türkiye’s digital transformation journey and working with all our strength to lead the vision of becoming a technology-producing and exporting country.
With the momentum continuing in our pioneering lines of business, we made a solid start to 2024 in line with the forecast set we shared for the whole year,” said Türk Telekom CEO Umit Önal.
He emphasized that the year began positively, with revenue growth meeting expectations and the EBITDA margin exceeding projections. Despite intense competition in the market, mobile services continued to drive consolidated revenue growth.
He underscored that fixed broadband also expanded its contribution, aligning with the anticipated acceleration in average revenue per user (ARPU) growth, and the company is taking necessary actions to further enhance this segment.
Emphasizing that they play a locomotive role in Türkiye’s digital transformation journey, Önal said the company is weaving fiber networks across every corner of the country with the motto of "high-speed accessible internet for everyone.”
"We continue our investments with great motivation and dedication in the field of fiber, which is one of the indispensables for the technologies of tomorrow.
As of the first quarter of the year, our fiber network length, which reached 441,000 kilometers, covers 32.2 million households. Our fiber subscriber base reached 13 million in the first quarter of 2024, while our FTTC (Fiber to the Cabinet) subscriber base was 8.6 million and our FTTH/B (Fiber to the Home/Building) base increased to 4.5 million,” he noted.