Tag Archives: AI

AI Impact on Power and Broadband

AI technology seems to be a hot topic in every industry, and broadband is no exception. It seems inevitable that AI will be used to help monitor and control complex broadband networks. It looks like the biggest ISPs are already phasing AI into the customer service process.

The big question that nobody seems to be able to answer is if AI will change the amount of broadband the average household uses. It’s not an easy question to answer. It’s a reasonable question to ask because I seem to read weekly how AI is going to affect the way we communicate, and that seems likely something that will involve broadband.

An easier question to answer is AI’s impact on U.S. power consumption. It’s clear, at least for now, that AI and cryptocurrencies are fueling the construction of a lot of new data centers. The International Energy Agency’s (IEA) report for 2024 predicts a big uptick in worldwide power demand coming from data centers. IEA estimates that worldwide data centers in 2022 used about 460 Terawatt Hours (TWh) of power and predicts that by 2026 demand from data centers will grow to between 620 and 1,050 TWh. That would be the equivalent of adding as much energy used annually by Sweden at the low end of the estimate or Germany at the top end. IEA says that data centers in the U.S. will grow from using 4% of generated powerin 2022 to 6% in 2026.

The impact on broadband usage is harder to pin down. AI will impact broadband usage is several ways. There is the middle-mile impact of supporting the many new AI data centers. There are two types of data going to and from data centers. There is traffic sent to and from users asking AI to respond to queries. That means a public-facing AI data center should be equally as busy as a data center that responds to Google searches today. The second big use of broadband comes from feeding the public-facing data centers with the massive amount of data needed to ’train’ the AI. A public AI data center imports piles of data scraped from websites and other sources. An AI data center will create a busy node on the Internet that will draw a lot of traffic.

The biggest uses of future AI will probably not be in the big public AI data centers, but from data centers dedicated to big corporate users. Big corporations like Bank of America are not going to use the big public AI data centers, but will create their own AI data center to crunch their own data. One can picture a large new stream of Internet traffic coming from the many branches of Bank of America to feed the company’s own AI data center.

Both kinds of data centers will create new demand for long-haul and middle-mile fiber networks. Companies are likely to place AI data centers close to the existing long-haul networks that carry traffic from city to city. Companies that operate fiber transport networks are expecting a lot of AI-related traffic. I remember seeing that one of the justifications for the recent upgrade in Zayo long-haul networks was to prepare for AI.

The impact on home broadband is harder to predict. It’s possible that AI will decrease the amount of bandwidth used at home. If I research a topic today to write a blog, I do a Google search and perhaps visit a half dozen websites looking for background information. If I instead ask an AI search engine to find what I need, I’m going to look at fewer web sites if I’m satisfied with the AI answer to my questions. A recent article in Scientific American suggests that Google might use 30 times more energy to answer my question using AI instead of its traditional search engine. But at my home computer, I will likely use less bandwidth to get the condensed response from the Google AI. Most of my interfaces with AI involve transmitting short questions to the AI cloud, and receiving relatively short responses.

This doesn’t mean that there won’t eventually be more data-intensive uses for AI in the home. It may be possible to use AI to create a truly smart home that takes care of our needs automatically. I’m still waiting for the big virtual input screen that floats in the air in front of me like I’ve been seeing in futuristic movies. But for now, for most users, it’s hard to think that AI will increase bandwidth usage at home.

ISPs and AI

One of the most common questions I’ve been asked lately is what I think the impact AI will have on the broadband industry.

All of the big ISPs in the industry have actively been pursuing the use of AI. For example, AT&T Labs says it is investigating the use of AI to optimize the customer experience and auto-heal the network. Comcast says that it is using AI to help process petabytes of data every day. Comcast also worked with Broadcom to develop the first broadband chip for nodes, amps, and modems that bring AI into the network. Verizon is working on an AI solution to improve the customer experience in its IVR systems for customers calling the company. Charter is working AI into its customer interface. It’s also using AI to help customers generate commercials for advertising on the cable network.

Before talking about those uses, a basic primer on AI is needed. Most people are familiar with public AI platforms like Chat GPT or the Google Cloud Platform. No big corporations are using the open public versions of AI. Any data dumped into those systems is available to other users. Instead, corporations are buying and implementing private versions of AI that they train using their own data. One of the common issues with public AI platforms is that AI will hallucinate and invent an answer to a question. However, hallucination can be controlled in private networks where the user strictly controls the data.

All of the big ISPs, and seemingly most companies that field a lot of calls from customers, want to use AI to improve the customer experience. There are different approaches to using AI. One of the primary uses of AI is to eliminate customer menus where customers are asked to wade through a menu to choose who they want to talk to. AI can be used to interpret a customer request and direct the call to the appropriate place. AI can also be used to quickly pull all information about a caller to put it at the fingertips of a customer service rep. Maybe the most important feature of AI is that a customer conversation can carry across different customer service reps, meaning that a caller doesn’t have to repeat basic information every time they are transferred.

There are companies in the country that have completely automated AI to fully handle the customer interface, but it’s not likely that any big ISPs have gotten that bold yet. All of the feedback I’ve heard is that it’s still far too easy for an AI system to badly misinterpret what a customer wants. The same goes for attempts to fully automate an online chatbot. It doesn’t seem like anybody has come close to perfecting this yet, and doing it clumsily is frustrating for customers. But who knows, maybe in the future, most customer interfaces could be entirely handled by an AI representative.

Big ISPs are all investigating the use of AI in the network. The most obvious uses of AI is to interpret real-time network data to detect problems and analyze network quality. For many years, networks have used alarms to identify problems. One of the issues with an alarm system is that ISPs get constantly hit with minor alarms, and it’s not always easy to pick out the ones that matter. One of the hopes with AI is to look deeper at the performance of network equipment to identify problems long before an alarm is triggered.

ISPs are also starting to use AI for load balancing. It’s easy to think of broadband usage on a network as a steady state, but the reality is that usage spikes and dives erratically from second to second. AI can be used to examine usage on all segments of a network. For example, there are numerous paths from the network core in a fiber or cable network, and AI can examine all of them in real-time, as well as understand how usage spikes from neighborhoods can overwhelm other parts of the network.

The big temptation is to let AI take an active role in fixing problems. That idea makes a lot of network engineers nervous because AI is still nothing more than a series of algorithms created by programmers. It’s incredibly challenging for any programmer to create perfect programming, and the fear is having a network get out of control in a way that humans will have a hard time regaining control without shutting the network down. It’s not hard to envision an automated AI repeatedly magnifying and compounding a network problem.

The last use of AI by ISPs is to automate functions done by people. None of the big ISPs are talking about this because doing so sparks a lot of anxiety in the workforce. AI seems to be efficient at processing repetitive data or generating routine reports for management. It’s becoming obvious that other industries like banking and insurance companies have already been able to reduce some staff due to AI efficiencies. It’s likely that ISPs are already quietly reducing some clerical and middle-management staff due to AI. This is the part of AI that makes workers nervous. AI is more likely to replace white-collar workers and middle management than hands-on technicians. But this is going to be done quietly, at least until one of the big ISP CEO spills the beans on an investor call.

It’s going to be a while until any of these benefits move downhill to smaller companies. AI hardware and software is prohibitively expensive and smaller ISPs will have to wait until there are generic solutions offered by AI vendors.

The Future of Broadband Maps

I read that an AI expert at a workshop hosted by the FCC and the U.S. National Science Foundation suggested that AI could be used to produce better broadband maps. I had to chuckle at that idea.

The primary reason for my amusement is that FCC maps are created from self-reported broadband coverage and speeds by the many ISPs in the country. ISPs have a variety of motivations for how and why they report data to the FCC. Some ISPs try to report accurate speeds and coverage. People may be surprised by this, but some of the biggest telcos, like CenturyLink and Frontier, seem to have gotten better at reporting DSL speeds – in some markets, you can find DSL capability being reported at a dozen different speeds to reflect that DSL speeds vary by the distance from the central office.

Other ISPs take the exact opposite approach and report marketing speeds that are far in excess of the capability of the technology being deployed. It’s not hard to find WISPs claiming 100 Mbps to 300 Mbps download capability when they are delivering speeds in the 10 Mbps to 30 Mbps range. My guess is that some of these ISPs are using the FCC maps as an advertisement to get customers to call them after looking at the FCC map. Some ISPs have already been accused of over-reporting speeds to try to block grant money from overbuilding them.

There are also endless examples of ISPs reporting coverage that doesn’t exist. The FCC mapping rules say that only locations that can be served within ten business days should be included in broadband coverage areas, and many ISPs are claiming much larger areas than they can serve quickly. Even worse, some ISPs claim coverage in areas that they can’t serve, such as when WISPs claim coverage of homes that are blocked from line-of-sight by hills or other impediments.

The only way that AI could be used to improve the maps is if the FCC gets serious about mapping and changes some rules, and enforces others. The FCC would have to eliminate the ability of ISPs to claim marketing speeds, which provides easy cover for overstating capabilities. The FCC would also have to get serious about enforcing coverage to meet the 10-day installation rule. If those two changes were made and enforced, the FCC might be able to use AI to improve the maps. AI could match claimed ISP coverage to speed test data and also reference and compare coverage to complaints and challenges from consumers. I don’t see the FCC ever being willing to get that aggressive with ISPs – because this process would be extremely contentious.

I don’t believe any of this will ever happen because after the wave of BEAD funding is finally spent, the FCC and everybody else is going to lose interest in the broadband maps. Nobody will care if some ISP overstates capabilities in an area as long as the BEAD winner is going to bring faster broadband.

There are already a number of State Broadband offices that are saying that the BEAD allocations are not going to be enough to fix broadband everywhere. My prediction is that states that care about fixing the remaining places will create their own broadband maps and will go back to ignoring the FCC maps.

The FCC won’t care. At the point where they can say with a straight face that 95% of homes will be be able to buy broadband that meets the FCC’s definition of broadband, the FCC is going to declare job done. For the last decade, the FCC has issued annual broadband reports to Congress that have said that the state of broadband is good and improving – all based upon maps that everybody knew were grossly overstated in both broadband speeds and coverage. I can’t see the FCC putting extra effort into proving that there are still homes left without good broadband.

There is No Artificial Intelligence

It seems like most new technology today comes with a lot of hype. Just a few years ago, the press was full of predictions that we’d be awash with Internet of Thing sensors that would transform the way we live. We’ve heard similar claims for technologies like virtual reality, block chain, and self-driving cars. I’ve written a lot about the massive hype surrounding 5G – in my way of measuring things, there isn’t any 5G in the world yet, but the cellular carriers are loudly proclaiming its everywhere.

The other technology with a hype that nearly equals 5G is artificial intelligence. I see articles every day talking about the ways that artificial intelligence is already changing our world, with predictions about the big changes on the horizon due to AI. A majority of large corporations claim to now be using AI. Unfortunately, this is all hype and there is no artificial intelligence today, just like there is not yet any 5G.

It’s easy to understand what real 5G will be like – it will include the many innovations embedded in the 5G specifications like frequency slicing and dynamic spectrum sharing. We’ll finally have 5G when a half dozen new 5G technologies are on my phone. Defining artificial intelligence is harder because there is no specification for AI. Artificial intelligence will be here when a computer can solve problems in much the way that humans do. Our brains evaluate available data on hand to see if we know enough to solve a problem. If not, we seek the additional data we need. Our brains can consider data from disparate and unrelated sources to solve problems. There is no computer today that is within a light-year of that ability – there are not yet any computers that can ask for specific additional data needed to solve a problem. An AI computer doesn’t need to be self-aware – it just has to be able to ask the questions and seek the right data needed to solve a given problem.

We use computer tools today that get labeled as artificial intelligence such as complex algorithms, machine learning, and deep learning. We’ve paired these techniques with faster and larger computers (such as in data centers) to quickly process vast amounts of data.

One of the techniques we think of artificial intelligence is nothing more than using brute force to process large amounts of data. This is how IBM’s Deep Blue works. It can produce impressive results and shocked the world in 1997 when the computer was able to beat Garry Kasparov, the world chess champion. Since then, the IBM Watson system has beat the best Jeopardy players and is being used to diagnose illnesses. These computers achieve their results through processing vast amounts of data quickly. A chess computer can consider huge numbers of possible moves and put a value on the ones with the best outcome. The Jeopardy computer had massive databases of human knowledge available like Wikipedia and Google search – it looks up the answer to a question faster than a human mind can pull it out of memory.

Much of what is thought of as AI today uses machine learning. Perhaps the easiest way to describe machine learning is with an example. Machine learning uses complex algorithms to analyze and rank data. Netflix uses machine learning to suggest shows that it thinks a given customer will like. Netflix knows what a viewer has already watched. Netflix also knows what millions of others who watch the same shows seem to like, and it looks at what those millions of others watched to make a recommendation. The algorithm is far from perfect because the data set of what any individual viewer has watched is small. I know in my case, I look at the shows recommended for my wife and see all sorts of shows that interest me, but which I am not offered. This highlights one of the problems of machine learning – it can easily be biased and draw wrong conclusions instead of right ones. Netflix’s suggestion algorithm can become a self-fulfilling prophecy unless a viewer makes the effort to look outside of the recommended shows – the more a viewer watches what is suggested, the more they are pigeonholed into a specific type of content.

Deep learning is a form of machine learning that can produce better results by passing data through multiple algorithms. For example, there are numerous forms of English spoken around the world. A customer service bot can begin each conversation in standard English, and then use layered algorithms to analyze the speaker’s dialect to switch to more closely match a given speaker.

I’m not implying that today’s techniques are not worthwhile. They are being used to create numerous automated applications that could not be done otherwise. However, almost every algorithm-based technique in use today will become instantly obsolete when a real AI is created.

I’ve read several experts that predict that we are only a few years away from an AI desert – meaning that we will have milked about all that can be had out of machine learning and deep learning. Developments with those techniques are not leading towards a breakthrough to real AI – machine learning is not part of the evolutionary path to AI. At least for today, both AI and 5G are largely non-existent, and the things passed off as these two technologies are pale versions of the real thing.

AI, Machine Learning and Deep Learning

Data CenterIt’s getting hard to read tech articles any more that don’t mention artificial intelligence, machine learning or deep learning. It’s also obvious to me that many casual writers of technology articles don’t understand the differences and they frequently interchange the terms. So today I’ll take a shot at explaining the three terms.

Artificial intelligence (AI) is the overall field of working to create machines that carry out tasks in a way that humans think of as smart. The field has been around for a long time and twenty years ago I had an office on a floor shared by one of the early companies that was looking at AI.

AI has been in the press a lot in the last decade. For example, IBM used its Deep Blue supercomputer to beat the world’s chess champion. It really didn’t do this with anything we would classify as intelligence. It instead used the speed of a supercomputer to look forward a dozen moves and was able to rank options by looking for moves that produced the lowest number of possible ‘bad’ outcomes. But the program was not all that different than chess software that ran on PCs – it was just a lot faster and used the brute force of computing power to simulate intelligence.

Machine learning is a subset of AI that provides computers with the ability to learn without programming them for a specific task. The Deep Blue computer used a complex algorithm that told it exactly how to rank chess moves. But with machine language the goal is to write code that allows computers to interpret data and to learn from their errors to improve whatever task they are doing.

Machine learning is enabled by the use of neural network software. This is a set of algorithms that are loosely modeled after the human brain and that are designed to recognize patterns. Recognizing patterns is one of the most important ways that people interact with the world. We learn early in life what a ‘table’ is, and over time we can recognize a whole lot of different objects that also can be called tables, and we can do this quickly.

What makes machine learning so useful is that feedback can be used to inform the computer when it makes a mistake, and the pattern recognition software can incorporate that feedback into future tasks. It is this feedback capability that lets computers learn complex tasks quickly and to constantly improve performance.

One of the earliest examples of machine language I can recall is the music classification system used by Pandora. With Pandora you can create a radio station to play music that is similar to a given artist, but even more interestingly you can create a radio station that plays music similar to a given song. The Pandora algorithm, which they call the Music Genome Project, ‘listens’ to music and identifies patterns in the music in terms of 450 musical attributes like melody, harmony, rhythm, composition, etc. It can then quickly find songs that have the most similar genome.

Deep learning is the newest field of artificial intelligence and is best described as the cutting-edge subset of machine learning. Deep learning applies big data techniques to machine learning to enable software to analyze huge databases. Deep learning can help make sense out of immense amounts of data. For example, Google might use machine learning to interpret and classify all of the pictures its search engine finds on the web. This enables Google to be able to show you a huge number of pictures of tables or any other object upon request.

Pattern recognition doesn’t have to just be visual. It can include video, written words, speech, or raw data of any kind. I just read about a good example of deep learning last week. A computer was provided with huge library of videos of people talking along with the soundtracks and was asked to learn what people were saying just by how people moved their lips. The computer would make its best guess and then compare its guess to the soundtrack. With this feedback the computer quickly mastered lip reading and is now outperforming experienced human lip readers. The computer that can do this is still not ‘smart’ but it can become incredibly proficient at certain tasks and people interpret this as intelligence.

Most of the promises from AI are now coming from deep learning. It’s the basis for self-driving cars that learn to get better all of the time. It’s the basis of the computer I read about a few months ago that is developing new medicines on its own. It’s the underlying basis for the big cloud-based personal assistants like Apple’s Siri and Amazon’s Alexa. It’s going to be the underlying technology for computer programs that start tackling white collar work functions now done by people.