The title is pretty self-explanatory; I will be going over the technology that we have already or are currently developing that was made available thanks to the development of AI (other than general consumers directly using the LLM as a general assistant, since I assume you already know about that). Then I will go over what sort of technology could be available to us in the future.
On top of utilising AI to make AI based technology in itself, it is thought that AI will also exponentially speed up the process of innovation, leading to non-AI creations much faster. This in itself will be transformative for society if true. With the amount of progress science has made even in the past century, if AI speeds up this already exponential curve, who knows what the world could look like in another 100 years!
This blog will be focusing on current and future technology that directly uses AI, rather than what the world may look like from AI speeding up innovation in itself but thought it was worth noting!
Perhaps the most significant innovation as a consequence of AI outside of LLMs in their raw forms, AlphaFold is a program that is very useful for research into protein structures, and specifically for accelerating drug discovery! AlphaFold 2, released in July 2021, ended up winning a Nobel prize for its innovators Demis Hassabis and John Jumper from Google DeepMind in 2024. There was a long-standing unsolved problem relating to how protein structures unfold themselves. AlphaFold allowed scientists to use a program to give an approximation of how any given protein unfolds based on the protein structure. The reason this is important for drug discovery is the process of unfolding of proteins is a fundamental step to critical cellular processes that play a crucial role in how proteins interact within the body. When you want to predict what a drug is going to do in the body, understanding this step is crucial.
AlphaFold is the first time AI was able to solve an unsolved problem while at the same time genuinely improving science and medicine! Not only did this display the current value of AI but makes you wonder what AI will be capable of doing as the technology improves!
A very recent development, and one that definitely has a long way to go, but AI generated coding has been a big development. Cursor and Claude Code are two examples. This is particularly helpful for people who do not know how to code on their own. By explaining in English (no programming) the specification of the software you want built, these “coding agents” (agent being a term of AI that can do tasks autonomously rather than just output responses) then create the program for you. These agents are also much faster at coding than a human, being able to produce hundreds of lines of code in comfortably less than a minute.
These programs can also be linked to any relevant datasets and code to give code that is tailored to your specific pre-existing repository of code. For example, it can clean data from a chosen dataset and even perform analysis on a linked dataset.
A big limitation of this technology at the moment is the coding it generates can be riddled with bugs. For already technical people using this to improve their efficiency in coding, you turn from a coder to a debugger, having to check everything the coding agent is outputting is doing what it should. The amount of time saved varies greatly depending on how accurate the code is, and sometimes even how understandable the agent makes the code. On top of this, the code tends to lack conciseness, giving round-about solutions to problems. Not only does this make the code harder to debug, but it generally makes the entire codebase less readable and understandable relative to if a human had coded it themselves. Stacking on this added complexity could arguably make the use of coding agents slower when you look long term, particularly from a tech-debt standpoint.
These agents also currently struggle to produce good code with big datasets when you link them, which limits its usefulness for large companies. Snooping around online, experienced software engineers seem to be less impressed by this technology, but on the flip side, this opens a lot of doors for people who don’t know how to code!
While the future of technology using AI largely depends on the ability for the technology of AI to improve, there are current projects in development and ideas of where it could go. I will list some of them now.
Our ability to create drugs that cure/mitigate certain diseases and ailments revolve around our ability to observe, model and edit the human body at the microscopic scale. AI could potentially not only improve our ability to make innovations on our ability to observe, model and edit, but also it can be used in technology previously not possible. Neuralink is one example of this, but the possibilities are huge. AI could accelerate drug discovery massively, as it already has with AlphaFold.
In the novel version of “2001: A Space Odyssey”, the narrator makes a comment on a species much more developed than human beings, commenting they could “store knowledge in the structure of space itself, and to preserve their thoughts for eternity in frozen lattices of light” and that “They could become creatures of radiation, free at last from the tyranny of matter”. Maybe we use lattices of electrons passing through semiconductors rather than lattices of light to transmit the data of these digital minds, but this could maybe be in the rough ballpark of future reality!
Following on from biohacking, this is a brain-computer interface that allows for utilizing the neural signals in our brains to cause actions to physical objects in the real world. For example, prosthetic limbs being controlled using this technology. The success of this company would be revolutionary. For people who have been paralyzed and had their lives ruined, this technology could bring back their lives. Though this technology has already worked on patients/customers, it is still in clinical trials and is not available to the public yet. If this comes to the general consumer, which I expect it will, this would be a brilliant piece of technology. The only issue is the price. Especially when it is first released, this sort of technology doesn’t exactly sound like it’s cheap. This could be an extremely unfortunate way to widen quality of life between life brackets (a minority, but still people none the less). Perhaps fundraisers could be the way forward to mitigate this issue.
Another interesting direction with this technology is, rather than to revive human ability, to enhance it. I believe in a decade this is going to be a very big and heated topic of debate. Even on the Neuralink website, it says “This technology will restore autonomy to those with unmet medical needs and unlock new dimensions of human potential”. That last part strongly hints to the future vision of this company.
In the novel “Klara and the Sun” by Kazuo Ishiguro, which is set in the future where you can buy robots to have as a friend, there’s a very interesting concept in the book called “lifting”. Lifting is a surgical intervention done during early childhood to enhance a child’s intellectual capabilities but is a risky surgery that can leave children extremely ill or even lead to death. If you don’t “lift” your child, they are immediately put at an intellectual disadvantage and so are limited on choice of schools and careers and so will be limited in their material success in life. The main character “Josie” was “lifted” and suffers serious consequences as a result of it. Maybe this book was a look into the future and this concepting of “lifting” and the debates that come with it will be recreated by us when talking about our children.
Self-driving cars are also a very new innovation. While they have not spread across the world, they are in use in very specific places such as Los Angeles. Self-driving cars have the benefit of being able to drive “optimally” much more precisely than a human could. For example, they will be able to optimise their driving speed and lane changes to minimize stop-start traffic as well as inefficient braking and acceleration. With small optimizations everywhere, the hope is that on aggregate there is enough impact to free up road space without needing to build new roads. For the same reasons, this could also reduce energy consumption which would save money and help the environment.
This also allows people who cannot or struggle to drive to get around in a car. For example, the elderly and the disabled. It also gives the rider more freedom since they don’t need to focus on the road anymore. They can relax or be productive without being pre-occupied by focusing on the road. If this technology does spread and become popular as a taxi service, it could become much cheaper than buying your own car. These cars can also run 24/7! The downside is the potential unemployment this could cause as these are completely employee free, and taxi companies are incentivized by the lower costs in multiple areas as mentioned previously, but also the saved money from not needing drivers.
Both Google and Elon Musk have the goal of moving data centres from Earth to space. The idea is to get these data centres to orbit the earth in a way such that it is always facing the Sun. Then, you attach solar panels to these drifting data centres so that it can absorb the Sun’s radiation 24/7, rather than being limited to daytime. The amount of sun light received will also be much higher as the energy does not need to travel through the atmosphere, causing partial dissipation. On top of this, since this is outside the atmosphere and only using the Sun, this will remove the negative effects these data centres have on the environment as well as local communities.
The idea is that, as the technology and methodology improve, at some point in the future it will actually be cheaper to send data centres and compute to space compared to having them on Earth. Google CEO Sundar Pichai said "there's no doubt to me that a decade or so away we'll be viewing it as a more normal way to build data centers." Elon Musk said on Dwarkesh’s podcast this will happen in 3 years, and Jeff Bezos said space-based data centers will be operating in 20 years.
This is very ambitious and this claim has been put into question given severe logistical constraints. Mainly, the replacement and maintenance of these data centres and the infrastructure it contains. If you need to replace some number of GPUs, you have to send a rocket to do so. For maintenance, you either again need to routinely send people up there to carry out maintenance or you need people living isolated in space (personally I’d want a pretty good pay check for that, again making this expensive!)
Given there are different players with this aim, in my eyes this does give the idea more credibility since it suggests that multiple people, all of whom are presumably well informed on relevant topics, have all come to this same conclusion. Or at the very least, one agreed when the other publicly stated it. It’s definitely an interesting idea and if they can get it to work, it would definitely be a positive step! Who knows, maybe we can generate all our power using this method of space if we find fast and cheap way to maintain and repair technology we have sent to space.
While the field of creating human like robots with physical bodies has been going on for a while, as you would expect, LLMs really drove progress in this field. Much like the autonomous vehicles, these robots could work for us 24/7, never tire out, and if the technology gets good enough maybe even do tasks faster than humans can depending on the exact task. Again, like with autonomously driven vehicles, a downside is this could lead to unemployment in the field of labour. The replaceability of a labour job depends on the dexterity needed to do the job. The more dexterity needed, the better the technology needs to be before it will be used by companies.
Using robotics, maybe we could have surgeries and any other general medical procedure done by robots. This would be especially beneficial in countries with a shortage of talent, of which the majority of developing countries. If we can produce enough such robots, people who used to struggle to even eat could be receiving an extremely high level of medical care! This does depend on our ability to get AI robots to, not only do something incredibly precise like a surgery, but with a very high success rate. Just like with a human, a mistake could be detrimental to the patient, to the family, and expensive for those liable when the families take the AI robot manufacturer/hospital to court!
With AI increasing the ability to output, this led to a lot of benefits for developing countries. For example, we could have AI powered robots that can build infrastructure like hospitals, schools and homes extremely quickly. The increased supply would then also drive down the prices of these buildings. Likewise, we could do the same for mass production of food, meaning more widespread availability and again, this increased in supply would decrease its price.
Having these LLMs as useful assistants also lowers the barrier of entry to learning. While the internet allowed anyone with a device that could connect to the internet to have the vast amount of knowledge uploaded to the internet at their fingertips, LLMs lower the barrier to having a teacher that can ad-hoc explain and clarify anything you want! Anyone (with access to the internet, which to be fair still is not everyone) that has the drive and talent can acquire the theoretical knowledge they want! Of course, anything that requires hands on practice, which is most things, can not be taught using a text predictor. But hopefully with the previously made point, AI can also help with the development of the required infrastructure whether that be directly through lowering prices and availability by AI making infrastructure easier to build, or indirectly through promoting GDP growth via AI technology which then allows these poorer countries to buy what is necessary for their citizens to develop and learn.
One big hurdle with this is unfortunately corruption. If we look at certain countries in Africa, the destitution is not helped by severe corruption that limits the wealth and spending trickling down to the average citizen. With this, even if we were to grow GDP significantly or decrease prices of goods that would have high ROI for quality of life for average citizens in these countries, if the government just pocket it, the effects will be disappointing. Perhaps philanthropy from the potentially enormous wealth among those who greatly benefit from AI could help if the government won’t.
One of the scarier technologies AI has made possible, the range of weaponry available to be used for defence and warfare will change using AI.
One good use would be replacing human beings for war. If we just use robots to fight wars, this could remove the extremely tragic reality of mass human loss. On the flip side, if these robots are used to kill people rather than other robots, this could cause a lot of trouble. Potentially these robots would be much more effective if air-borne, and this greatly removes restrictions on their ability of movement, in particular to population dense areas. This could make it very easy for countries to suddenly invade and very hard for countries to defend themselves if attacked suddenly. Whether this stops more battles and wars than it starts all depends on whether AI is misused, which in turn depends on the people using it, and the legal and technical guardrails implemented in these devices/robots.
AI could also lead to the creation of weapons that don’t already exist, or enhancement to pre-existing weapons. Sound, smell, and light sensors that detect the presence of life beyond human capabilities for example. There are allegations of Google creating missiles sold to Israel to use on the Gaza strip that utilise AI to detect sound to identify any hiding people.
Another terrifying thought is biological weapons. The development of viruses that are designed to optimise for being contagious and fatal. While a government is unlikely to do this, there are concerns of random people using AI to get the steps necessary to carry out such attacks. This is essential to guardrail against in AI but in the case that doesn’t work, a successful attack could harm a lot of people very quickly, potentially too quickly for governments to act. Though covid did prep us for mass research and mass production of vaccinations quickly, and even if governments anticipate such attempted attacks and prepare, some amount of damage is bound to be done and at that point, all we can do is try and mitigate the numbers of those impacted.
There have also been concerns around the US government following in given the disagreement with red tape put down by Anthropic around their deal with the US government, specifically with autonomous weapons and mass surveillance (the latter will be talked about after weaponry) and wanting the power to use this technology for “all lawful uses” as stated in the contract written by the US government and sent to Anthropic. I will cover this in a later blog, but what is considered lawful could be bent extremely far given the current lack of legislation on the uses of AI. As of writing this, just a few days ago the US government were blocked from trying to impose sanctions on Anthropic for not agreeing to the demands of the government and pulling out of the deal. This sort of bullying tactics over Anthropic trying to prevent misuse from the government, while it failed this time, is genuinely concerning.
Ultimately, while the potential for weapons is unsettling, a country is required to develop weaponry and other mechanisms for national interest / defence (not offence!). If a given country does not make them, you are at a big risk of being bullied by more aggressive countries who do. It could even be similar to the leverage gained by a country becoming a nuclear power.
Another one of the more concerning technologies, the abilities of mass surveillance will be enormous with the development and adoption of AI. This is particularly a concern with authoritarian regimes such as China, where substantially developed mass surveillance is already in place.
With the use of AI, the capabilities of mass surveillance grow massively. Any government has the potential to monitor, store and track massive amounts of information once the correct surveillance infrastructure was in place. In particularly extreme scenarios, George Orwell’s 1984 could turn into a prophecy, with every single one of your moves being recorded. While this is unlikely to happen in the west, it is still quite the scary prospect.
As mentioned with weaponry, mass surveillance was also a concern that Anthropic had around their deal with the US government. I would imagine nothing to the extent that China almost certainly will be deployed in the US, but still quite frightening! Particularly with the president constantly trying to push the boundaries on what he is allowed to do, this is a well needed reminder that misuse is very much possible in democratic nations!
“Closing the loop on AI” by creating AI that can make AI opens up the possibilities for efficiency gains allowed by AI. On top of improving the efficiency that AI can be created, potentially (when AI is of a high enough standard) it will be able to make innovations allows them to create AI that humans would not have conceived of, or at least that would have taken a much longer time. This would then hopefully create a snowball effect, where AI accelerates the amount AI can produce efficiency gains for humans! Potentially this can get us to the point where improving labour efficiency and rate of innovation is no longer the bottleneck for the rate of output, particularly if AI continues to improve as it has been.
This could also really make a difference in quality of AI from country to country. Those who have managed to “close the loop” see this exponential of exponentials in AI development. This could lead to tremendous amounts of leverage over other countries if even just a few years ahead given this speed of development. This however banks on the method of “closing the loop” not being leaked/having a barrier to entry to execute in practice which is not guaranteed.
Overall, there are lots of potentially life changing technologies which AI could unlock, along with extremely dangerous technology. AI very much looks to be a double-edged sword, and whether any self-inflicted wounds to humanity are lethal to its own existence depends on the shrewdness and sensibility of a very small portion of society. If not lethal, the possibilities are endless! The world as we know it is expected to look completely different, even unrecognisable 10-20 years down the line. Who knows, maybe human beings will too.
With AI integrated across society, jobs looking completely different, cars driving around without a human driver, potentially even robots walking among the streets with humans as our assistants, labourers, cleaners etc, more bionic limbs among humans, maybe energy that is completely derived from space, extremely advanced abilities to fight diseases, countries using robot based warfare in the news, AI being significantly smarter than all humans by a significant margin and many other differences, kids born after these transformations will look at our current lifestyle as we look at people in the 1800s. Primitive, unrelatable, shocking and vastly inferior!