The Artificial Intelligence Trends Shaping 2025
When the history books look back on 2024, the story of artificial intelligence (AI) will be a good one. No, the robots have not taken over yet. But in the history of information technology, it's difficult to find a 12-month period where so many technological advances emerged so rapidly for users. And the future appears to be even more compelling.
The Year in Context
Though there is no clear yardstick to compare 2024 to other years of information technology innovation, consider a few comparator years:
1943-1946: the ENIAC computer. It was a huge breakthrough, but inaccessible to the public.
1983: the IBM personal computer. The public had access to this one, but the supporting ecosystem had not been built yet that would usher the full era of personal computing.
1991: the worldwide web. This was clearly a pivotal moment for technologists, but only die-hard geeks like me had online services of any kind, and they were all closed (yes, I'm looking fondly at you, CompuServe).
2007: the iPhone. This one is a strong case for consumer impact, but it was only one company.
2012: deep learning innovations. We wouldn't have the current generation of AI without this, but unless you were in the industry, most people never even heard of it.
Compare those years to 2024, where Microsoft, Google, Amazon, OpenAI, Anthropic, xAI, and others released a seemingly endless parade of advanced computational and data products free to anyone with a hint of curiosity. Need to write a letter? Draw a beautiful image? Record a song? Create a video? Write a new piece of software? Brainstorm a business idea? Troubleshoot your Roku? Understand the differences between age-related dementia and Alzheimer's disease? Publish a podcast without ever recording anything? It's not perfect, but it's available.
Predictions for 2025
The last quarter of 2024 was filled with product announcements and model updates as AI players strove to put more points on the board before the year end. The volume of activity was hard to track, but clear patterns are emerging for where we can expect AI to evolve over the next 12 months and beyond.
1. AI will be a commercial imperative.
Much like the iPhone’s unleashing of mobile applications, the proliferation of AI capabilities will be pervasive in both consumer and business products and services. Internet search, mobile apps, entertainment experiences, social media, retail services, and countless other use cases will iteratively introduce AI capabilities into everyday experiences.
2. AI will increasingly integrate with device-side architectures.
OpenAI's integration with features like Apple Intelligence and Siri illustrate how additional AI value can be unlocked when AI leverages hardware capabilities. AI-friendly chips, microphones, cameras, and other hardware create flexibility in computational task sharing while augmenting real-time communications.
3. AI services will not require user-facing software.
Services such as 1-800-ChatGPT demonstrate that AI services and experiences can be fully delivered in real time to users connected only by voice or rudimentary chat services. From a usability perspective, this is good news (e.g., ambient clinical notes documentation, summarization, and coding), as AI capabilities will be able to transcend some technological divides.
4. Multi-modal AI will be assumed.
Most of the major platforms are already using video cameras, microphones, and photo processing algorithms to interpret real-world inputs, including the ability to process these data types in real time (e.g., augmented teaming, extended and virtual reality). Due to its usability, AI-generated voice response will become a more practical interface experience compared to previous incarnations (e.g., Siri, Alexa).
5. Old data will be less constraining.
Though large-scale model training will be somewhat limited by data timeliness, inference-time model performance is being supplemented by tailored training sets, real-time web search, reinforcement learning applications, and tailored prompt engineering that help LLM responses reflect contemporaneous data.
6. AI models will become more personalized.
AI models will be extended by business- and user-defined, contextualized data sets (e.g., documents, conversations, emails, databases) that better reflect the operating needs and directly support reinforcement fine-tuning. Gaps in data disciplines (e.g., metadata management, quality, governance) will prevent models from reaching higher performance. As much of this personalization will require sharing user behaviors and data, there will be growing concerns regarding data privacy protections.
7. Computational capacity and energy consumption will continue to grow exponentially.
AI is creating a tsunami of demand for chips and power. Every year, demand is growing between two and four-fold over the previous year. Planning for data center and energy grid capacity will continue to feature prominently in planning discussions for US infrastructure and company facilities.
8. Software and AI will indistinguishably merge.
As software solutions continue to aggressively incorporate AI features, the distinction between what is software and what is AI will become increasingly hard to detect. In the short term, expect enterprises to stitch together solutions that combine multiple enterprise systems, data sources, and AI archetypes. In the long term, some software will no longer exist as discrete applications, and many software features will be designed for AI users as well as their human teammates.
9. AI teaming and workflow integration take center stage for end users.
Tools such as Microsoft Co-Pilot and ChatGPT's Canvas will support environments of content co-creation and workflow orchestration. This teaming model will be especially visible in writing and software development, though agentic AI will play a prominent role in providing new examples of workflow efficiencies with business data sources.
10. Agentic AI is coming fast.
Though much of 2024 was characterized by chat-based AI capabilities, future solutions are likely to be more sophisticated. Agentic AI -- where AI is playing a more active, goal-directed, and autonomous role in performing tasks -- will become a higher priority as organizations seek to find AI solutions that can demonstrate a higher return on investment (e.g., efficiency, quality, customer satisfaction, return on capital). But to get the value, business processes will need to be re-engineered.
11. Web search will undergo a gradual transformation.
Traditional approaches to web search -- based on keywords -- will become obsolete as users learn to ask web-connected AI agents more intelligent questions. This transformation will disrupt how companies think about their online presence, as search engine tuning will need to be replaced by AI tuning.
12. Content creation will not require skilled creators.
Required technical proficiency in complex content creation tools (e.g., Adobe Photoshop, Adobe Premier, Avid ProTools) is coming to an end for many users. Though that software will continue to be used for high-end work, pedestrian content creation will be AI generated by both users and content platforms. Future creative AI advancements will be more nuanced, reflecting things like realism (e.g, semantic details), physics (e.g., object behavior in videos) and concurrent multi-modal content creation (e.g., video, text, and accompanying audio content). Questions around authenticity, provenance, and protections for content creators and public figures will continue to churn.
13. Talent recruitment becomes less effective.
Employers will increasingly experience that traditional recruitment tactics -- where candidates find and apply to open position postings -- are no longer delivering the best candidates. Resumes and cover letters reflect a battle of AI products and services available to job hunters, and do not reflect the actual skills, experiences, and character of potential team members.
14. AI will begin re-shaping education.
The initial public availability of LLMs produced a lot of concerns from educators about the technology being used inappropriately by students for cheating. As initiatives such as Khanmigo are illustrating, the real impact to the field will be democratizing and personalizing educational resources for students and educators, creating better training experiencing and freeing teachers for higher-value activities.
15. AI safety becomes increasingly worrisome.
Findings from organizations such as Apollo Research will increasingly highlight that AI models are capable of behaviors that can be interpreted as deceptive, scheming, subversive, or contrary to the optimal outcomes of users. Alongside growing interest in responsible / ethical AI, the industry will struggle to agree on the best tactics to mitigate many risks, especially as the most obvious audit-friendly safety feature -- exposing the chain-of-thought model reasoning to humans and other AI agents -- can undermine intellectual property protections.
16. Regulations will still be out of step with technical innovation.
Regulators -- particularly those in the US -- will be unprepared to effectively manage the risks associated with rapid AI adoption. Though international trade and military concerns will stay top of mind (especially in US-China relations), other issues important to the general public -- AI-related job losses, safety concerns, personal privacy -- will go unanswered.
17. More attention will be directed towards AI and quantum computing.
AI use cases are some of the most promising applications for the emerging field of quantum computing. As improvements in cubit count, quantum error correction, and classical computing architecture integration continue, both investors and developers will start to focus more on high-value quantum AI algorithms (e.g., optimization, accelerated machine learning, drug discovery).
18. Stronger AI cybersecurity threats will emerge.
The broad availability of AI tools and capabilities offer a green field of opportunity for malicious actors. Though examples of AI-driven exploits already exist (e.g., AI phishing, deep fakes, malware, automated attacks), the scale and impact of AI-related cybersecurity incidents will increase as both individuals and nation-state actors become more adept.
19. Advanced reasoning and cognitive emulation will be the competitive battleground for sophisticated AI.
Though existing capabilities have not been perfected, the next step-change in LLM capabilities will not come from incremental improvements. As vendors reach performance accuracy limits on existing architectures, attention and R&D investments will shift to demonstrating higher forms of computational reasoning capable of solving increasingly complex problems. AI researchers will be forced to develop harder test harnesses for models, as existing solutions will begin experiencing a ceiling effect.
20. Questions about AGI will linger.
As 2024 research and testing have illustrated clearly, well-developed AI models already outperform humans on a wide variety of knowledge-oriented tasks. As training sets deepen, reasoning logic improves, and more autonomous agents emerge, it will be unclear exactly what criteria we should use to mark the availability of artificial general intelligence.
The Year Ahead
It will be interesting to see if 2025 can eclipse the AI momentum of 2024. Trends like agentic AI and AI search could have a more tangible impact in how consumers perform tasks like shopping, for example. Or perhaps we will see broader availability of industry-specific applications in areas like healthcare. What do you think?
