The global forex market is the largest and most liquid financial market in the world, operating twenty-four hours a day, five days a week. Traders face constant price movements influenced by factors such as central bank policy, macroeconomic releases, and geopolitical events. Managing such complexity requires speed and clarity, and that is where artificial intelligence (AI) tools can be helpful. AI is not replacing human judgment, but it can be used a companion for processing vast amounts of data, detecting subtle patterns, and executing trades with discipline.
Forex trading presents a unique challenge compared to equities or futures. Trading is active 24/5 and liquidity is very deep for major currency pairs. Using high leverage is very common. Traditional methods (manual chart analysis, news monitoring, and rule-based strategies) are often insufficient in capturing the nuances of currency behavior. AI tools can be used to expand the trader’s reach by integrating multiple inputs, from technical indicators to news sentiment, and adapting to market shifts. Machine learning systems, for example, can analyze how EUR/USD responds not just to a rate announcement but also to prior market positioning and cross-asset movements. These insights are difficult to capture manually.
Examples of How AI Tools Can Be Utilized by Forex Traders
Market Scanning and Pattern Recognition
One of the most practical applications of AI in forex trading is pattern detection. Chart patterns such as double tops, flags, or head and shoulders are traditionally spotted by eye, but AI models can scan thousands of charts across multiple timeframes and identify recurring structures with greater accuracy.
AI-based scanners can also adapt to volatility conditions. A moving average crossover in a calm market may carry different weight compared to the same signal during a high-volatility session. By training on historical data, AI can filter setups that are statistically stronger, saving traders from false signals that clutter traditional scanning methods. As always, be aware of the risk of overfitting when using historical price data.
Sentiment and News Analysis
Forex is particularly sensitive to macroeconomic events, such as interest rate decisions, employment figures, inflation releases, and geopolitical developments. AI-powered sentiment analysis tools parse central bank statements, economic reports, and even social media chatter to gauge market tone. Natural language processing models extract sentiment indicators that traders can overlay on charts. For instance, if sentiment around the US dollar shifts sharply negative after an ambiguous Federal Reserve statement, an AI sentiment tool may detect this earlier than traditional methods. Swing traders and intraday traders alike can benefit from this added perspective, combining technical setups with a real-time measure of mood in the market.
Algorithmic Execution and Automation
Automation has long been part of forex trading and retail forex traders have for instance used the so-called Expert Advisors (EAs) available for the major trading platforms MetaTrader 4 (MT4) and MetaTrader 5 (MT5). AI takes automation further by allowing models to adapt strategies in real time. Reinforcement learning systems, for instance, adjust position sizes or stop levels dynamically as market conditions evolve.
Some AI systems execute directly through broker APIs, splitting orders intelligently across liquidity providers to reduce slippage. For traders running high-frequency or scalping strategies, these execution improvements can materially change outcomes. Even for swing traders, automation reduces the risk of emotional decision-making when trades must be managed overnight.
Please observer that the CMA have regulated the use of Algos in forex trading.
Risk Management
AI tools are increasingly used to monitor account-level exposure in forex. Because leverage is so central to currency trading, managing risk dynamically is critical. AI-driven risk engines calculate exposure not only pair by pair but also across correlated currencies. A trader long EUR/USD and long GBP/USD may unknowingly be doubling risk to the US dollar. AI can flag this correlation in real time and suggest adjusted sizing.
Stress testing with AI goes further by modeling scenarios. What happens if the US releases unexpectedly strong inflation data? How might that cascade across USD, emerging market currencies, and safe havens like JPY and CHF? By running scenario models, AI tools prepare traders for possible outcomes before they occur.
Backtesting and Strategy Refinement
Forex traders have long relied on backtesting strategies against historical data. AI enhances this by automating parameter optimization and identifying market regimes where strategies succeed or fail. Rather than testing a moving average crossover blindly, AI examines how that crossover performs under different volatility conditions, across sessions, and against multiple pairs. By recognizing when a strategy works best, AI helps traders avoid overgeneralizing results. Instead of deploying a strategy across all conditions, traders can target specific environments where probability of success is higher.
Accessibility of AI in Forex
Access to AI is no longer limited to institutional traders. Retail traders can access AI features directly in trading platforms, either through native tools or third-party integrations. TradingView, MetaTrader 5, and cTrader all host custom AI-driven indicators or signal services. Independent platforms such as Capitalise.ai allow traders to automate strategies in plain language, with AI handling the coding. For those with programming skills, Python frameworks such as TensorFlow and PyTorch enable building custom AI systems, connected to broker APIs. Cloud computing services further reduce barriers by allowing AI models to be run without expensive local hardware. A trader can build a model on a laptop and deploy it on a cloud server to trade continuously, freeing resources and reducing downtime risk.
While AI has not completely transformed mainstream retail trading platforms, it is increasingly appearing as built-in features, third-party plug-ins, and integrations. The large retail trading platforms MetaTrader 4, MetaTrader 5, and cTrader all incorporate AI, mostly through AI tools from the surrounding developer ecosystems. While MT4, MT5, and cTrader do not yet have AI embedded as a standard feature, they have become host environments for AI-driven tools through their open architectures. MT4 offers the largest marketplace, MT5 provides the strongest optimization and multi-asset support, and cTrader delivers the most flexible integration for developers comfortable with mainstream programming. As AI becomes more common in trading, these three platforms are expected to expand their official support, but for now, traders rely heavily on third-party plug-ins and integrations to bring AI into their workflow.
Across all major forex platforms, AI adoption is driven less by the platforms themselves and more by the ecosystems around them. Marketplaces are filled with indicators, expert advisors, and scripts that are using AI or are claiming to use AI. Some are basic adaptive algorithms while others genuinely incorporate machine learning. The most common features AI plug-ins provide include:
Adaptive indicators that adjust automatically to changing volatility.
Pattern recognition tools trained to identify formations like flags, head and shoulders, or triangles.
Sentiment overlays drawing from news feeds or social media.
Risk management modules that optimize position sizing dynamically.
Signal generators that use trained models to issue buy or sell recommendations.
Integration with Python and other machine learning environments is also becoming more common. Traders can build AI models externally using TensorFlow, PyTorch, or scikit-learn, and link them to their trading platforms for execution.
MetaTrader 4 (MT4)
MetaTrader 4 is the most established forex trading platform, and while it was released long before the current wave of AI adoption, its popularity and open architecture have allowed third-party developers to bolt on AI capabilities. The platform itself does not contain native AI features, but expert advisors coded in the MQL4 language are widely available. Many of these use machine learning principles, such as adaptive indicators that adjust parameters in real time or pattern recognition tools trained on historical data. Plug-ins marketed as “AI advisors” often use neural network models to generate signals for currency pairs. There are also third-party services that connect external AI systems to MT4 accounts, typically through APIs. These may run sentiment analysis in the cloud and feed back trading signals into the platform. While the quality of such add-ons varies, the sheer size of the MT4 marketplace means that AI-driven tools are widely offered, even if not officially part of the core platform.
MetaTrader 5 (MT5)
MetaTrader 5 was built as a successor to MT4 and offers a more modern foundation for AI integration. Its programming language, MQL5, is more powerful and closer to mainstream coding standards, making it easier to implement advanced logic. Like MT4, most AI capabilities in MT5 come from expert advisors and plug-ins rather than native features. However, MT5 supports more complex data handling, which benefits AI systems that require larger inputs. Developers have created neural network-based advisors, adaptive trading systems, and sentiment-driven signal generators. Another difference is the stronger integration of backtesting and optimization. MT5 allows more robust testing of AI-driven systems across multiple cores and cloud networks. This makes it more practical for traders building or licensing AI models that need to be validated on large historical datasets.
Examples of MT4 / MT5 AI Add-Ons & Integrations
The Auto-GPT MetaTrader Plugin. This is an open-source project that connects a MetaTrader account (MT4 or MT5) to Auto-GPT. The purpose is to allow Auto-GPT (a generative AI agent) to fetch market data, place or modify trades, and respond to instructions given in natural language. It’s not turnkey for all traders and setting it up involves connecting via APIs, configuring permissions, and understanding how bots read signals. Traders can benefit from the flexibility, but must verify safety and avoid over-automation without oversight.
Acuity / SmartFX Signal Centre Plugin. This is an AI-powered plugin for MT4/MT5 offered by SmartFX / Acuity. The “Signal Centre” adds premium signals, combines financial/market data, and sometimes sentiment or event-based alerts. It enhances standard charting with trade ideas, economic calendar integration, and attempts to merge human & AI insight. Useful for swing traders who want trade ideas and alert-based assistance, rather than full automation. The customization is limited, so results depend heavily on how well the signals align with the trader’s style.
cTrader
The trading platform cTrader approaches AI differently. Its algorithmic trading environment, cAlgo, is based on the C# language, which is a widely known and used programming language. This makes cAlgo more accessible for developers familiar with mainstream programming and it is easier to connect with external AI libraries written in .NET. Traders can use third-party machine learning frameworks to build AI models and integrate them directly with cTrader for execution.
While the cTrader trading platform does not come with built-in AI features, the combination of C# support and open APIs makes it particularly attractive for those who want to experiment with AI models beyond pre-packaged indicators. Developers have created plug-ins that integrate cTrader with Python AI models, allowing traders to run deep learning or reinforcement learning systems externally and then feed signals into the platform.
Examples of cTrader AI Plug-ins and Add-Ons
DeepSeek Plugin v1.0. This is an AI-powered market intelligence plugin used inside cTrader. DeepSeek helps interpret indicators, chart patterns, and trends automatically. It also assists with refining cBots (automated strategies) by offering suggestions for optimization and debugging. It can generate alert-style market updates and summaries of recent events. This plugin is suited for traders who want decision support without coding. The AI helps interpret what the market is doing and highlights potential trades, but final decisions still rest with the trader.
CodePilot AI Assistant / cTrader CodePilot. CodePilot is a chat-style AI assistant built for cTrader. It allows traders to describe trading tools, indicators, or strategy logic in conversational language, and then generates corresponding C#-compatible code (cTrader’s automation language) for indicators, cBots, or strategies. For example, a trader can ask: “Generate an indicator that generates a buy when the 40-period MA crosses above the 100-period MA on EURUSD”, and CodePilot will produce a working indicator. This helps non-coders prototype ideas quickly.
WebView Plugins via cTrader Store. The cTrader plugin ecosystem includes WebView add-ons that embed live data, charts, sentiment summaries, and sometimes AI-assisted tools available directly from the cTrader store. These may not always be labeled “AI,” but often include features like auto-pattern detection or dynamic alerts.
Comparisons & Trade-Offs
Real-time alerts vs backtesting: Some tools are better for generating alerts (DeepSeek, Signal Centre), while others like CodePilot and Auto-GPT allow backtesting or building custom indicators which can be tested historically.
Integration risk: When using AI-enabled tools that place trades automatically (Auto-GPT plugin, expert advisors), there’s risk if the model misinterprets data or market conditions change abruptly. Signal plugins that just alert tend to be safer, assuming the trader retains discipline.
Data sources & latency: Performance depends heavily on underlying data quality, refresh rate, and latency. A plugin that recommends trades based on stale data can mislead. Tools built into platforms like cTrader often have good real-time access; external AI systems might lag depending on feed.
Cost and transparency: Some AI add-ons are paid and proprietary, with limited visibility into how their models are built or how accurate they are historically. Others (like open-source or community plugins) may provide more transparency but possibly less polish or support.
Dangers of AI Tools in Forex Trading
Artificial intelligence is being marketed as a powerful force in forex trading, with claims that it can parse news, detect chart patterns, and even generate profitable trading signals. Independent testing of AI systems, however, reveals that they are often far less reliable than in the marketing material. For forex traders, who commonly use a lot of leverage, the risks of over-trusting AI are magnified. Recent evaluations of leading AI systems highlight not only the risks of depending on them but also the specific weaknesses that traders need to understand. Independent testing shows that many models deliver confident but unreliable signals. AI tools often misinterpret economic events or fabricate data when asked for specifics outside their scope. Traders must guard against over-reliance, while also building discipline into their workflow, treating AI as decision support rather than as an autopilot.
AI-generated insights should ideally be verified against independent sources, but this can of course be difficult to find time for in situations where every second counts.
Another issue that traders need to be aware of is the risk of overfitting. Overfitting is not unique to AI; it can happen any time when a trader fits a strategy so well to historic price data that it will struggle to handle future price movements. Forex markets evolve, and a system that thrives in one environment may break down in another.
Strengths and Weaknesses for Six Different AI Platforms
When tested for their ability to analyze financial markets and provide actionable insights, each AI platform has displayed certain strengths, weak spots, and limitations. An issue shared by all of them is that they tend to deliver outputs with an air of certainty regardless of reliability. This is dangerous in trading, where many traders already struggle with a psychological tendency to believe in signals that confirm their bias. When AI delivers flawed analysis in confident prose, a trader will happily believe it, provided that it feels right. Instead of questioning the output, traders may defer to the machine, and end up losing money.
ChatGPT
The strength of ChatGPT lies in its ability to produce detailed explanations and structured strategies. It can describe technical setups, explain macroeconomic concepts, and generate trading plans with clear logic. The problem is that these plans can seem very reliable while masking deeper weaknesses. ChatGPT struggles with live market data, as it does not have direct access to real-time prices or order flow. In testing, ChatGPT has occasionally produced fabricated numbers when asked for specifics about company earnings or currency rates. For forex traders, where intraday shifts can invalidate a strategy within minutes, the reliance on invented or stale data is a direct hazard.
ChatGPT is an AI-powered conversation agent developed by OpenAI, and it generates human-like responses to text-based input. ChatGPT is based on a machine learning model called a transformer, specifically the GPT (Generative Pre-trained Transformer) architecture, and it was trained on a vast amount of internet text. During training, ChatGPT learned language patterns by analyzing huge datasets from books, websites, and more. From that base, it has been refined through RLHF (Reinforcement Learning from Human Feedback).
As of 2025, the two main version of ChatGPT are the free ChatGPT-3.5 and the for-pay ChatGPT Plus GTP-4-turbo. The GPT-4-turbo is faster and more capable, but you must pay a monthly 20 USD subscription fee to use it. The free version does not store your information between sessions. The paid version is faster, can handle longer conversation, can resolve more complex tasks correctly, and is more widely used professionally.
Gemini
Gemini was developed with a focus on speed and the ability to summarize, and it typically produces results faster than its peers. In testing, Gemini has excelled at condensing complex information into digestible overviews. In forex, where news events like employment data or central bank statements require quick interpretation, this responsiveness can be advantageous, but Gemini also showed gaps in financial interpretation, often mislabeling neutral policy announcements as bullish or bearish signals. In essence, we get higher speed at the cost of accuracy. A forex trader relying on oversimplifications from Gemini can easily end up positioned against the market.
Gemini is not just one tool; it is the name for a whole suit of artificial intelligence models developed by Google Deep Mind. Launched in 2024, the Gemini suit succeeded Bard AI. The Gemini models are good at processing and integrating multiple data types, such as text, image, audio, and video.
The Gemini 2.5 series include models capable of reasoning through tasks step-by-step, leading to more accurate and reliable outputs. The Gemini 2.5 Flash model is especially capable when it comes to scanning and analyzing extensive documents and answering complex queries. From a forex trading perspective, Gemini’s ability to process and integrate diverse data types and handle vast amounts of information can be helpful when it comes to analyzing market news, sentiment, and technical indicators simultaneously.
As Gemini is a product from Google, it is now integrated into the web browser Chrome, making it easily accessible to Chrome users. Gemini is also in the process of being rolled out to professional users through Google Workspace.
Developers can integrate Gemini’s API with trading platforms to automate trading strategies, although this requires technical expertise. Many vendors are offering various short-cuts, but trader caution is warranted. Just as with many other type of AI, there are scammers and low-quality providers who use the word gemini in their own branding and marketing in an effort to make their products seem more trustworthy. It’s important to distinguish between Google’s Gemini AI and other platforms that may use similar branding. There is for instance the GeminiPRO Trader, which is not a Google product. The GeminiPRO Trader claims to offer AI-powered trading signals and high-accuracy bots, but transparency is lacking. Another example is GiniTrade AI, marketed as a neural AI-driven trading system with a 99% accuracy rate. Yes, wow, 99%. While we have not personally tried the GiniTrade AI, outrageous claims like that are typically warning signs.
Claude
Claude distinguishes itself with a clear, readable tone and well-structured narratives. It provides cleaner answers than some of its peers, which can reduce confusion for traders interpreting its output. However, it also struggles when it comes to handling numerical data and context-sensitive analysis. In testing, Claude sometimes failed to grasp the nuance of events such as central bank forward guidance, where subtle wording shifts matter more than headlines. For forex traders who depend on interpreting nuance in policy language, this weakness can create a false sense of clarity while missing the actual driver of price movement.
Claude AI is a family of advanced large language models developed by Anthropic, an AI research company founded by former OpenAI employees. Claude is being marketed as an AI with an emphasis on safety and ethical considerations, where the aim is to provide helpful, harmless, and honest interactions. Anthropic utilizes a special training model they call Constitutional AI, where human-written principles are employed to guide the model´s behavior, promoting safety and consistency.
As of 2025, the most recent version of Claude is Claude 3, which includes models such as Haiku (optimized for speed), Sonnet (balanced performance), and Opus (designed for complex reasoning tasks). These models can process both text and images, and demonstrate strong abilities for logical analysis and programming across various languages and frameworks.
Team and Enterprise users can benefit from Claude´s memory feature, where user-specific details are stored. (It can be turned off in settings.) All users, including those using the free version of Claude, can run the AI in incognito mode to increase privacy.
Perplexity
Perplexity is designed as a research-oriented AI assistant, and has received praise for linking to its sources when answers are generated. This transparency appeals to traders who are accustomed to verifying data. The problem for traders is that Perplexity often relies on general financial news or educational articles rather than raw market data. For forex traders, where a one-minute delay on a central bank announcement can change the trade setup, it is a major weak spot. Perplexity is marketed as a an AI-powered search engine with real-time web integration, but this does not mean that Perplexity´s is always using the most current data available.
Perplexity AI is a United States based company founded in 2022. The Perplexity AI search engine uses
language models (LLMs) combined with real time web search. Perplexity AI does not create its own foundational models, but instead uses models from other AI companies such as OpenAI, Antropic, Google Deep Mind, Mistral, Meta, and xAI. Whatever issues are present in the model or models used can impact the accuracy of Perplexity´s answer.
In testing, Perplexity sometimes provided summaries that were factually correct in general but irrelevant to the immediate market context. This creates a subtle danger: traders may feel reassured by the citations while still acting on signals that are too slow or too broad for profitable forex trading.
There is a free public version and more than one for-pay premium versions available. Perplexity Pro subscribers can access several backend models, including GPT-5, GPT-4.1, Claude 4.0, and Gemini Pro 2.5, offering flexibility in response generation.
In a recent study where Perplexity AI was asked for bibliographic reference retrieval, it generated a significant number of erroneous or fabricated references, and this problem with fabricated information is something forex traders should take into account as well before putting too much trust into Perplexity AI.
Grok
Grok AI was developed by Elon Musk’s company xAI and is designed for inference and real-time data processing, integrating closely with Elon Musk’s ecosystem, including the X platform. While Grok models, such as Grok 3 and Grok 4, have demonstrated strong performance in various benchmarks, including finance, they are not without limitations and weak spots.
In tests, Grok 3 Mini Fast Beta (High Reasoning) scored an average accuracy of nearly 82%. Grok has also faced criticism for its occasional hallucinations, and for being over-reliant on real-time data from platforms like X, which may lead to the inclusion of unverified or heavily biased information in its responses. This issue is concerning for traders, since inaccurate and unreliable information can end up generating big losses. Some users have also reported inconsistent performance, with response times varying significantly, which could impact its reliability in time-sensitive trading situations.
While Grok AI offers impressive speed and can access real-time data, it shares some of the same challenges as other AI systems are struggling with when it comes to interpreting nuanced financial events.
Designed with a touch of wit and a rebellious streak, Grok’s responses are often characterized by humor and a conversational tone. While this approach aims to make interactions more engaging, it can be an issue for traders who would prefer a more serious and fact-focused approach.
To access Grok, you need an X Premium+ subscription, which at the time of writing costs $16 per month. This subscription provides additional features and removes ads from the platform. Grok is available exclusively through the X platform and does not have a standalone app.
Meta AI
Meta AI enters with the weight of a large ecosystem, integrated into social platforms and designed for broad consumer accessibility. Its strength lies in usability and integration, providing answers in natural, conversational form across devices. For some forex traders, this availability and ease of use make Meta AI it the first AI they consult. The weakness is that Meta AI, like others, does not deliver reliable financial specificity. Testing showed that it often generalized and offered vague summaries of market conditions without actionable detail. In some cases, Meta AI exaggerated certainty about rate expectations and price levels. This is hazardous in forex trading, where leverage magnifies even small errors. Meta AI’s broad reach among retail traders risks normalizing casual use of AI outputs in trading decisions, which can lead inexperienced traders to take risks without proper confirmation.
Meta AI is the artificial intelligence division of Meta Platforms, Inc. (formerly Facebook, Inc.), established in 2013 as Facebook Artificial Intelligence Research (FAIR). In 2021, the division was rebranded as Meta AI. Meta AI has developed several different AI tools, including LLaMA Language Models, AI Studio, Meta AI Assistant, SAM 2, and Audiobox. At the time of writing, the most recent version of the LLaMA (Large Language Model Meta AI) is the LlaMA 3.1, and it is available in several sizes, including the Llama 3.1 405B model. With 405 million parameters, this is one of the most capable of the publicly-available large language models. It can handle very long inputs without forgetting earlier parts, and it has performed great when tested for reasoning, long-form text summarization, generation of synthetic data, and coding. To run well, it needs very substantial computational resources, including powerful processors.
AI Strengths and Weak Spots – And the Implications for Forex Traders
The testing underscores that AI is better suited for background assistance than for generating direct trading signals. It can explain macroeconomic terms, outline risk management techniques, or provide examples of technical strategies, but it cannot replace live analysis and human judgment. The lack of access to up-to-date prices, the tendency to fabricate details, and the inconsistent accuracy across systems make AI unsuitable as a decision engine for leveraged forex positions.
For traders, the lesson is not to abandon AI but to recognize its limitations. ChatGPT can for instance be useful for structured strategy explanations, Gemini for fast summaries, and Claude when clarity of expression is of essence. None of them, however, deliver the precision, data integrity, or contextual sensitivity required to act as reliable forex advisors. Treating them as educational tools, not trading systems, is a safer approach.
AI tools are becoming part of the forex trader’s toolkit, offering advantages in market scanning, sentiment detection, risk management, and execution. They allow traders to analyze more pairs, process more data, and refine strategies with greater precision. Yet they also carry dangers when treated as infallible. The best use of AI in forex is as a partner: a system that extends human reach but remains subject to human judgment. The responsibility for interpreting signals, managing risk, and maintaining discipline continues to rest with the trader.
Understanding the Background
The artificial intelligence (AI) field is looking extremely promising right now and we are likely to see some huge improvements in the nearby future, including improvements that can be very useful for forex traders.
With that said, it is important to not get carried away by this latest hype. As traders, we need to remain calm, and carefully evaluate the different AI tools we encounter, making sure we understand both their abilities and their limitations.
Since the field av modern AI research was established in the 1950s, AI has gone through more than one cycle of boom and bust. Looking closer, we see a fascinating story of over-promising, under-delivering, and periodic “AI winters”, especially in terms of research funding and commercial investment. A boom period is typically driven by technological breakthroughs, and is followed by a bust period marked by disillusionment when real-world results fail to match the lofty expectations.
Post-WWII: The First AI Boom
The first boom began to build in the 1950s. World War II was over, the zeitgeist was optimistic, and the mathematical genius Alan Turing had just published “Computing Machinery and Intelligence” (1950), introducing the Turing Test. The researcher John McCarthy coined the term “Artificial Intelligence” in 1956 and believed (alongside many other experts in the field) that human-level AI would be achieved in a few decades. Programs doing symbolic reasoning and playing games such as chess and checkers looked very promising, and investment grants began to pour in from government, especially the U.S. government which financed a lot of research through the Defense Advanced Research Projects Agency (DARPA, a research and development agency of the United States Department of Defense responsible for the development of emerging technologies for use by the military.
The 1970s: The First AI Winter
The first so-called AI Winter, a period marketed be decreased interest in AI development, commenced in the early 1970s. The Lighthill Report, commissioned by the UK government, was published in 1973 and concluded that AI had made minimal progress and still had very limited practical applications. The UK government cut almost all government funding for AI research, except for a few expert systems.
Soon, AI projects in other countries, especially the United States, were also losing funding, and we embarked on a period characterized by low interest in AI, and strong feelings of disappointment due to unmet expectations. Early AI researchers made bold claims about creating machines that could understand language, reason, and learn like humans. But the actual systems developed (like early natural language processors and logic programs) were limited, brittle, and not scalable. Part of the problem was that existing hardware was still to slow and expensive to meet the needs of ambitious AI programs. Funding agencies and academic institutions began to shift focus to other areas, like traditional computer science and operations research. The first AI Winter thought AI researchers to be more cautious with their projections.
The 1980s: A Strong But Short-Lived Second Wave of AI Hype
The 1980s brought a new wave of AI hype, with the rise of Expert Systems like XCON (used by DEC to configure computer systems). These systems could use rules and logic well, but only in narrow domains.
In the 1980s, a lot of investments in AI came from the private sector, including corporate investments and venture capital. The Japanese government also invested heavily in AI through its Fifth Generation Computer Project.
The Second AI Winter
The second AI Winter started in the late 1980s and continued throughout the rest of the century and into the early 00s. The expert systems developed in the 1980s had turned out to be brittle, difficult to scale, and prohibitively expensive for commercial use. The could not learn and they failed to adapt to new scenarios. A lot of companies who had invested heavily in AI went bankrupt, and Japan´s Fifth Generation project was scrapped.
During the Second AI Winter, AI was almost a taboo word in tech and research funding circles, but some more narrower fields were still attracting some attention, including machine learning, computer vision, and Natural Language Processing (NLP). Instead of talking about artificial intelligence, projects in need of funding would use terms such as “autonomous agents” or “intelligent systems”.
AI in the 21st Century
The field of AI began to slowly recover and garner some more interest in the early 21st century. Computer processors became cheaper and better, Big Data became a thing, and neural network research became fashionable once more.
A Deep Learning Boom started in the early 2010s, with massive progress in Natural Language Processing (NLP), a field of AI focused on enabling computers to understand, interpret, and generate human language. Bit Tech invested billions, and we saw results in the form av advanced email spam filters, voice assistants such as Alexa and Siri, and machine translation tools such as Google Translate.
In the 21st century, advancement in related fields have been instrumental to the development of AI tools. Computer hardware has become better and more affordable, and since the early 2010s, graphics processing units have been used to accelerate neural networks. We have also seen the collection of immense data sets thanks to the internet, and the transformer architecture that debuted in 2017 is now used to produce impressive generative AI applications.
The 2020s has so far been marked by extremely rapid progress in advanced generative AI and we are currently firmly in an AI boom. This had led some analysts to fear that we are quickly approaching another AI Winter.
This article was last updated on: October 15, 2025