The biggest developments in AI since the COVID-19 pandemic are generative AI chatbots built on large language models, AI-driven protein structure prediction, the first large-scale commercial robotaxi services, and growing use of machine learning in quantum computing research. The pandemic did not create these technologies, but it pushed work and services online, raised demand for automation, and showed how fast AI tools could help researchers handle huge data sets.
Below is a grounded look at what actually changed, with dates and examples you can check, and where the hype outran reality.
Quick timeline of key milestones
| Date | Milestone |
|---|---|
| February 2020 | Baidu applies its LinearFold algorithm to the novel coronavirus RNA and open-sources it for researchers |
| October 2020 | Waymo opens its fully driverless ride service to the general public in the Phoenix area |
| November 2020 | DeepMind’s AlphaFold 2 achieves breakthrough accuracy in the CASP14 protein structure assessment |
| November 30, 2022 | OpenAI releases ChatGPT, setting off the generative AI boom |
| October 2024 | Nobel Prize in Chemistry awarded to David Baker, Demis Hassabis and John Jumper for computational protein design and protein structure prediction |
| Q2 2025 | Baidu reports about 2.2 million fully driverless Apollo Go rides in a single quarter |
AI in vaccine and medical research
A popular claim is that AI “created” the COVID-19 vaccines. That overstates it. The first authorized vaccines, such as the Pfizer-BioNTech and Moderna shots, were built on mRNA technology that scientists had been developing for decades, and they still went through large human clinical trials. What AI and fast computing did was speed up parts of the research around them.
LinearFold and RNA structure prediction
RNA molecules fold into shapes that affect how they behave, and predicting that “secondary structure” is computationally expensive for long genomes. Baidu’s LinearFold algorithm scans a sequence in one pass and prunes unlikely options, which makes it run in roughly linear time. In February 2020, Baidu said applying LinearFold to the novel coronavirus cut prediction time for its genome from about 55 minutes to 27 seconds, and it made the tool available to researchers. Faster structure prediction helps scientists study a virus and evaluate candidate sequences, but it is one input among many, not a shortcut through clinical testing.
AlphaFold and protein structure
The more important long-term breakthrough is protein structure prediction. In late 2020, DeepMind’s AlphaFold 2 predicted protein shapes with accuracy close to laboratory methods in the CASP14 assessment, solving a problem biologists had worked on for about 50 years. DeepMind and EMBL-EBI later released a public database of predicted structures. The work was recognized with the 2024 Nobel Prize in Chemistry, shared by Demis Hassabis and John Jumper of Google DeepMind and David Baker of the University of Washington, whose lab designs new proteins computationally. These tools are now used in drug discovery, enzyme design and basic biology.
Self-driving cars and robotaxis
Autonomous vehicle research goes back well before the pandemic, and the idea that driverless cars became “the new normal” during COVID restrictions is not accurate. The real story is that a few companies moved from test programs to commercial service in the years that followed.
- Waymo opened its fully driverless service, with no safety driver behind the wheel, to the public in the Phoenix area in October 2020, and has since expanded to more US cities.
- Baidu Apollo Go runs robotaxi services in a number of Chinese cities, including Wuhan, and reported about 2.2 million fully driverless rides in the second quarter of 2025 alone.
These systems combine several kinds of AI: computer vision to recognize vehicles, cyclists and pedestrians from camera, lidar and radar data; prediction models that estimate what other road users will do next; and planning software that chooses a safe path. Deployment is still limited to mapped service areas, and every expansion depends on local regulators, which is why progress has been gradual rather than sudden.
Chatbots and generative AI
Customer service chatbots existed before 2020, but most followed rigid scripts and frustrated users who went off-script. The pandemic increased the volume of online support requests, and businesses looked for ways to handle them. The real turning point came with large language models.
Large language models are trained on large amounts of text to predict the next word in a sequence. At scale, that simple objective produces systems that can answer questions, summarize documents, draft emails and write code in conversational language. Research models such as Baidu’s ERNIE-GEN (2020) explored better ways to generate natural text, and OpenAI’s release of ChatGPT on November 30, 2022 brought the technology to the general public. Since then, competing assistants from Google, Anthropic, Meta and others, plus many open models, have made generative AI a standard business tool.
For companies, the practical change is that a support bot can now understand a wide range of phrasings, pull answers from a company’s own documents, and hand off to a human when needed. Teams new to AI chatbot development usually start with a narrow use case, such as order status or FAQs, and expand once accuracy is proven. Known limits still apply: language models can state wrong information confidently (often called hallucination), so businesses typically ground answers in verified data and keep humans reviewing sensitive cases.
Generative AI has also spread into marketing and media. Our article on how AI video ads affect consumer engagement covers one example, and the future of live streaming looks at AI’s role in real-time content.
AI and quantum computing
Quantum computers use qubits, which can exist in a superposition of 0 and 1, to tackle certain problems that are very hard for classical computers, such as simulating molecules. They are not general replacements for ordinary computers, and today’s machines are still small and error-prone.
AI and quantum research now support each other in two ways:
- Machine learning helps run quantum hardware, for example by improving calibration and by decoding errors in quantum error correction schemes.
- Quantum machine learning explores whether quantum circuits can speed up some learning tasks. Toolkits such as Baidu’s Paddle Quantum, launched in 2020, let researchers build and train quantum neural networks in simulation, and cloud platforms such as Baidu’s Quantum Leaf give access to quantum programming environments.
Practical, large-scale quantum advantage for everyday AI workloads has not arrived yet, so treat bold claims in this area with caution.
AI chips and infrastructure
Training and running large models requires specialized hardware, mainly GPUs and custom accelerators designed for the matrix math that neural networks rely on. Demand for this hardware grew sharply after the generative AI boom, driving huge investment in data centers and chip design by companies such as Nvidia, Google, Amazon and others. If you follow this side of the industry from an investing angle, our overview of top AI stocks explains the main players.
How the post-pandemic era changed AI adoption
- Remote and digital work created more digital data and more demand for tools that automate routine tasks.
- Health research showed the value of sharing data and computational tools quickly across borders.
- Consumer familiarity with AI jumped once chat assistants became free and easy to use.
- Regulation moved up the agenda, with new rules such as the EU AI Act and ongoing debates over safety, copyright and privacy.
Limits and open questions
Rapid progress has brought real concerns along with it. Language models can produce confident but false answers, so outputs used for health, legal or financial decisions still need human checking. Training data raises copyright and privacy disputes that courts and lawmakers are still working through. Large AI data centers use significant electricity and water, which has become part of the debate over where and how fast to build them. Robotaxis still face questions about how they handle rare situations and who is responsible after a crash. Keeping these trade-offs in view is the best way to judge new AI announcements on their merits rather than on headlines.
Frequently asked questions
Did AI create the COVID-19 vaccines?
No. The leading vaccines used mRNA technology developed over decades and went through human clinical trials. AI tools such as Baidu’s LinearFold helped speed up some research steps, like predicting viral RNA structure.
What is the biggest AI breakthrough since the pandemic?
Two stand out: large language models, brought to the public by ChatGPT in November 2022, and AlphaFold’s protein structure prediction, which earned a share of the 2024 Nobel Prize in Chemistry.
Are self-driving taxis available to the public?
Yes, in limited areas. Waymo opened fully driverless rides to the public in Phoenix in October 2020 and has expanded since, and Baidu’s Apollo Go operates robotaxis in several Chinese cities.
How does AI help quantum computing?
Machine learning helps calibrate quantum hardware and decode errors, and toolkits like Paddle Quantum let researchers train quantum neural networks. Large practical gains for everyday AI have not arrived yet.
