Artificial Intelligence – or AI – has grown from a fringe idea in 1950s computer science to a household term used across popular culture, science and technology.
Once just a futuristic concept, AI has now penetrated wider society and is embedded in consumer technology, e.g. in the form of Siri, Alexa, Cortana and Google Assistant to become part of our daily lives. AI now also plays a key role in data analysis, marketing, advertising, medicine, science and engineering where computers are learning from stimuli and reacting in ways more human than ever before.
AI guesses what you’re going to write into search forms or messages, correcting your mistakes, helps track your orders with online retailers, powers map services like Google Maps and protects you against financial fraud and security breach. AI devices diagnose life-threatening medical conditions, calculates financial risks with enormous precision and can even help guide our own educational learning.
AI lives not just in machines and devices but behind internet browsers and within wireless networks.
Here, we’ll be reviewing the pros and cons of AI both in the present and in the future.
But What is AI?
Simply defined, AI are computerised systems that are in some way able to mimic human thought processes and perform similar tasks to us or other intelligent/sentient beings. They can interpret and respond to stimuli by ‘thinking’ as we can.
AI may use sensory data and machine learning techniques to collect and interpret information from external stimuli. In short, then, AI needs to incorporate some form of learnt behaviour. A typical consumer computer is not AI; despite its ability to compute information usually beyond our own intelligent capability, it is not able to ‘think’ in a meaningful enough way to be considered ‘intelligent’. True AI is hard to define because thinking is hard to define. The Turing Test is one such method of inquiry that can be used to discover whether a technology is capable of thought.
AI has always been a fascinating subject for philosophers, psychologists, science fiction writers and futurists as it poses the tantalising question; what happens when machines surpass the intelligence of their creators? From The Matrix to Terminator, Blade Runner and iRobot, AI is recurringly defined as a potentially uncontrollable force that can spiral out of humanity’s grasp if we’re not careful.
The Advantages of Artificial Intelligence
1. Efficiency and Accuracy
AI is scalable and efficient. Instead of being limited by the finite resource of a human brain, it can be integrated with scalable computer systems that range from cell phones to the supercomputers that power AI such as IBM Watson and Google DeepMind. The scalable efficiency of AI means it can be used to perform both very small tasks and enormously complex tasks over and over again without tiring.
It’s a misconception that all AI has to be ‘clever’ to be ‘intelligent’. AI can be used to conduct tasks that are otherwise very boring, time consuming or repetitive for humans, like crawling through webpages to collect information. Would you fancy trawling through a million webpages for pieces of information, manually copying and pasting them into a spreadsheet?!
2. Eliminate Human Error
AI can be trained out of ‘human error’ as they never lapse in concentration or are otherwise not adversely affected by environmental conditions. For example, a doctor who is tired may be prone to making poor decisions and tend to make more mistakes. AI is not vulnerable in the same way and remains relentlessly accurate. For example, Google DeepMind can now diagnose serious eye conditions as well as the world’s leading opticians and ophthalmologists and is even able to recommend the best kind of treatment in some 94% of cases.
This applies in a commercial context too. In warehousing, companies such as Amazon, Ocado and Walmart have started using coordinated teams of robots to sort, replenish and select stock. Ocado’s wireless robot control system uses a network of 1,000 robots which are issued commands some 10 times a second. This has eliminated human error.
3. Reducing Costs (Cheaper Products & Services)
AI systems that replace low-skilled human jobs, which means that your business might need to hire fewer people. Additionally, AI can then elevate people to focus on higher cognitive tasks that require more advanced training where a human can ultimately create more value.
This means that we will be able to improve the efficiency of our local, national and global businesses whilst reducing costs and making it cheaper for the end customer.
Businesses that fail to utilise AI might struggle to compete against innovative, highly efficient AI driven companies.
4. Improving Human Decision Making
Classification models which predict a label (is the result male or female?) given a set of input data can be incredibly powerful for businesses. Predicting useful labels on new customers allows you to make better decisions.
Several examples of this include:
- Customer churn models. (Will the customer churn or not?)
- Lead scoring models. (Will the lead be a good customer or not?)
- Content research (Will this article rank well in Google Search or not?)
5. Improving Human Workflows
In the field of natural language processing, it’s now possible to generate content briefs and first draft articles with natural language generation. This means that writers can spend more time editing and formatting blog posts rather than working on creating a first draft.
Additionally AI models can scan 1000s of content pieces and can help content writers and publishers to understand what words, topics and entities must be included in a good article to improve its relevance to the reader.
6. The Mechanical Advantage
Perhaps the thing that scares SciFi writers and futurists the most is that AI can be integrated into very powerful machines that have a mechanical advantage over humans. Imagine a machine with the strength of a tank combined with superhuman intelligence.
Military funding has always helped drive technological development. What was essentially an arms race between the US and Soviets turned out to be one of the major catalysts of space exploration – we are perhaps seeing this process mirrored with AI. China and the USA are potentially locked in an AI arms race, with both superpowers seeking to develop powerful autonomous combat robots for land, sea and air.
AI combat drones and vehicles are already in use, such as the Boeing Loyal Wingman. The use of AI allows engineers to push components beyond what a human could physically take whilst also not risking life in their operation. This is what we see in space also, where AI robots are able to boldly go where no man has gone before and collect data from inhospitable environments such as the surface of Mars and beyond.
In medicine, AI prosthetics are being built which can learn from the human nervous system to adapt in the same way as living tissue. By machine-learning from real nervous system data, prosthetic limbs are able to learn how to behave just like them. They can then be linked to our nervous systems to allow us to control them with thought alone, the same as we would a normal biological limb.
7. Effective Data Acquisition and Analysis
Computers have always worked with data extraordinarily well and AI is extremely good at working with high volumes of data that humans simply cannot handle. Not only can AI systems acquire and extract data at a phenomenal pace, but they can also interpret it and transform it, checking for any errors, inconsistencies, formatting issues, etc.
AI is replacing much of the manual work in data science and analysis, it can make decisions on our behalf, collecting relevant data based on relatively broad and abstract queries. AI can be used to locate insights and trends, say if you want to check what adverts had the highest conversion rates, what products certain customers tend to gravitate towards, their preferences for marketing, etc.
8. Understanding High-Dimensional Data
If you have data with say over 200 columns for each customer row, it can be incredibly difficult to perform customer segmentation with traditional marketing methods such as Recency, Frequency & Monetary Analysis.
However, using data science, we can leverage advanced dimensionality reduction techniques such as:
After reducing the dimensions to 3 – 5 columns, it’s possible to run a clustering algorithm on the data using either:
Finally, after running a clustering algorithm, you can view the relationships using either t-SNE or UMAP in a two-dimensional space.
By reducing your data’s dimensionality and clustering it, you can find hidden patterns and groups within it that would have been impossible to see without using unsupervised-machine learning.
This can be incredibly useful for:
- Advanced customer segmentation.
- Finding topic groups or similarities between many HTML web pages.
- Finding similar types of individuals with social media data.
The Cons of AI
There is credibility to the fears of AI as encapsulated by SciFi writers and echoed by those such as Hawking and Musk. Advanced AI is indeed a new horizon and we do not yet understand its limits and how careful we need to be in its development.
Whilst we may generally consider that an inventor always retains ultimate control over their invention, with AI, this may not be the case. We have already seen some evidence of unexpected AI behaviour, such as DeepMind exhibiting ‘aggressive’ behaviour during tests designed to assess how well it cooperated with others during stressful situations. Once AI becomes autonomous or liberated from its designer and highly intelligent, that is when Pandora’s Box is really opened. Though we might typically imagine this as a rogue intelligent robot that escapes the clutches of its inventor, intelligent rogue AI could ‘live’ anywhere with an internet connection – it doesn’t need to take on a physical form.
There are two likely scenarios where AI could become dangerous:
- AI is programmed to do something destructive
In the case of weapons development, AI could be programmed to kill, hurt or otherwise destroy people or property. Such devices would not be fielded with a simple means to deactivate them, or they wouldn’t be much good in combat. They’d likely be extremely hard to stop. Additionally, if they did use machine learning to learn from combat situations to improve their tactics or strategy, these learning processes could have unexpected and disastrous consequences.
- AI Becomes Destructive as a Side Effect
AI is generally designed to be relentless in its approach to solving logical issues. Humans are capable of nuance, so for example, we may judge that whilst rerouting a river may prevent flooding, it may also result in tremendous environmental destruction, thus the answer is to not take action at all.
AI built on the basic logical premise ‘to prevent flooding’ does not have the same liability or capability of nuance, and may go to destructive means to achieve its basic goal. This concept has been explored thoroughly in SciFi where machines once designed to help us turn against us as they begin to see that humans are in the way of the greater good.
2. AI Machines Don’t (Currently) Have Any Emotion
Perhaps then, the solution is to not invent extremely clever AI unless we can also teach it to be kind and compassionate? The problem is, these are still human ideas that might be interpreted differently by intelligent AI. Compassion and kindness are not entirely logical, but distinctly abstract and despite our best efforts to ‘design’ it, we will most likely always fall short of the mark.
Just because AI can be installed into a machine does not mean it can’t degrade. Human bodies are exceptionally good at regenerating and though we may consider flesh and blood a disadvantage over metal, there are some significant advantages. If the components in an AI system do degrade then unless it is somehow able to self-repair, it will not be able to fix itself and will slowly fail. It may be able to understand what is failing but is still physically powerless to prevent it or repair it. This may be the case where the model training data is now old or outdated.
4. A Reduced Number of Jobs For Humans
AI threatens our way of life. At the moment, the public’s appetite for AI is not very high and though we generally see little ‘evidence’ of AI in our day-to-day lives, this will change with automation. Automation threatens the labour market, we’ve seen how Ocado uses a network of robots to deal with warehousing and this scenario is proliferating rapidly potentially threatening millions of jobs worldwide.
Whilst it might seem a huge advantage to replace humans with more efficient robots, we have to consider where this ends. The Disney film WALL-E encapsulates this well, portraying humans as literal couch-potatoes who have over-consumed planet Earth. What happens to our identity when all of our jobs are gone? What will we do? Can it ever be sustainable?
The issue is not just sociocultural but political also. For example, in the UK, labour unions such as the CWU, who represents the Royal Mail, have already been heavily criticising automation plans that put people out of work.
5. High Costs
AI varies hugely in cost. If we consider a web scraper designed to scrape just hundreds of webpages then sure, we can run likely run this comfortably off our own internet connection and system resources. Scale this up to millions of webpages and suddenly, you require dedicated servers and powerful systems to run the web scraper efficiently. With AI’s scale comes scaled costs, too. It’s currently not within the grasp of most consumer technology to run complex AI.
Whilst simple AI is usually not power-hungry, it’s still expensive to run very complex AI. One of Google’s AlphaGo system used over 1000 CPUs and 200 GPUs! Not only are systems like this expensive and tough to build, but they’re also all physically large and not easily transported. This may change with quantum computing and other developments, but for now, super-powerful AI is expensive and cumbersome.
6. Lacking Creativity and Out-of-the-Box Thinking
As mentioned, AI is a logical beast. Numbers, symbols, words; anything data-related can be adeptly consumed and interpreted by AI. But what about abstract thinking and creativity?
Part of the beauty of the human mind is that it can be fairly volatile and non-linear in the way it works. This may also be a weakness as well, but humans at the cutting edge of knowledge have not been entirely logical or rational thinkers. We rely on our creativity to generate ideas to then explore logically. AI cannot currently do this and lacks the ability to think out-of-the-box. It can be taught set variables and learn to adapt them but this still occurs in-the-box – for now!
7. Ethical Considerations
AI may pose a great ethical challenge for humanity if it reaches a very advanced stage. At what point may a machine be deemed sentient, conscious, and therefore entitled to similar to what we call human rights? We may never reach this stage of AI – it may not be as hard as some imagine – but early awareness of AI’s ethical considerations is vital going forward.
Another ethical consideration is given the costs of advanced AI, it is generally monopolised by the world’s biggest businesses. Who owns the data collected by AI and what can companies legally do with it? DeepMind has already been embroiled in data controversy surrounding DeepMind Health and how it interacts with NHS records.
AI Use Cases
AI is used in a variety of technologies and can valuable for any process where human judgement, thought or decision-making is useful but where scale is important; AI can be used to complete millions of tasks in the same time period that humans can complete one.
AI in Commercial Settings – AI can be used to automatically mine and analyse data which can be implemented in commercial products and services. Machine learning algorithms combined with web scrapers can locate information automatically online, understand it and use it to guide pricing strategies, marketing campaigns and advertising. AI is frequently used by chatbots that have replaced human customer service agents for businesses worldwide. AI can also be used to measure prices and economic activity to forecast stock market trends and crashes.
AI in Automation and Engineering – AI can go where humans can’t. Space exploration is a key example. The AEGIS system developed by NASA and currently in use by the Mars rover can intelligently discern and measure information from Mars. Autonomous vehicles for use here on Earth are currently being tested by Tesla and Toyota amongst other manufacturers. In engineering and manufacturing, AI can be used to replace human labour, e.g. by learning the most efficient way to complete a task and replicating that on an industrial scale.
AI in Healthcare – AI can be used to calculate medical factors e.g. infection, transmission, prognosis and survival. It can learn from dynamic data to simulate medical scenarios, this is something we’ve seen in use lately with the coronavirus pandemic. AI combined with neural interfaces can provide amputees with synthetic limbs that function like the real thing or even help victims of brain injury or accident walk and talk again.
AI in Climate, Agriculture and Environment – Similarly, AI is being used to learn from enormous historical data sets that describe the climate, weather systems and global warming to measure and predict the progress of climate change. AI here can be used to predict tectonic activity and adverse weather events; hurricanes, flooding, tsunamis, etc.
AI’s Novel Uses – AI has any novel uses, e.g. MuseNet uses machine learning to learn about rhythm, melody, harmony and composition from ‘listening’ to millions of musical pieces. It can generate 4-minute compositions using 10 different instruments inspired by musical styles ranging from Bach to the Beatles and Metallica. Wordsmith can take huge quantities of data and inscribe it into written content that describes what data means so we don’t have to go through and interpret it all manually. It produced 1.5 billion pieces of content from petabytes of numerical data last year.
AI represents an element of our future, of that there is no doubt. It is sure to become a more prolific entity of our day-to-day, possibly even changing them with autonomous vehicles such as driverless cars, machines that do our work for us, analysing data, composing our music, possibly even writing articles such as this one without us even having to lift a finger.
It’s exciting for sure, but fear is warranted and though conceptions of AI taking over humanity are still confined to SciFi, we cannot be blind to the possibilities. For now, AI has served us pretty well and has made life easier in various ways. Where the balance needs to be struck is still not clear.