TikTok CEO Shou Zi Chew believes artificial intelligence could help deliver major scientific breakthroughs within the next two to three years, potentially changing how researchers work in physics, biotechnology, chemistry and materials science. Speaking at the DBS Asia Future Forward Conference in Singapore on October 12, 2026, Chew said the technology is still at an early stage of its development and that its long-term potential may be greater than many people currently expect. His comments highlight a growing shift in the AI industry toward scientific discovery, beyond consumer chatbots, content generation and productivity applications.
Chew’s prediction comes as ByteDance, TikTok’s parent company, expands its AI research through its Seed team and develops models for scientific applications. One example is Protenix-v2, a system designed for biomolecular structure prediction and design that could support drug discovery research. AI is already helping scientists analyze biological data, explore potential materials and accelerate computational experiments, but a major discovery still requires validation through scientific methods and, where necessary, laboratory testing. Chew’s two-to-three-year timeline is a forecast rather than a confirmed deadline for a specific breakthrough.
Key Takeaways
- Scientific breakthroughs by 2029: TikTok CEO Shou Zi Chew expects AI to contribute to major discoveries within the next two to three years.
- Four major fields: Chew highlighted physics, biotechnology, chemistry and materials science.
- ByteDance research: The company’s Seed team is developing AI systems for scientific applications, including Protenix-v2.
- Drug discovery potential: AI models can help researchers predict biomolecular structures and design candidate molecules.
- Faster scientific research: AI could help analyze complex datasets, identify patterns and narrow down promising experiments.
- Human validation remains essential: AI-generated hypotheses must be tested before they can be accepted as scientific findings.
- Safety concerns: Chew also warned about cybersecurity risks and the dangers of giving AI agents excessive permissions.
- Timeline is uncertain: The prediction reflects Chew’s outlook, not a guarantee that a breakthrough will occur within three years.
What Did TikTok CEO Shou Zi Chew Say About AI?
Chew made his remarks during a discussion about the future of AI at the inaugural DBS Asia Future Forward Conference in Singapore. The event brought together business leaders, investors and technology executives to examine how AI could change industries and economic activity.
Chew said he believes significant scientific advances could emerge in the next two to three years. He specifically identified physics, biotechnology, chemistry and materials science as areas where AI could help produce important results.
He also expressed a broader view about the pace of technological change, arguing that people may overestimate AI’s impact in the short term while underestimating its potential over a longer period.
That distinction is important. AI tools can generate impressive demonstrations, but integrating them into scientific research requires more than producing plausible answers. Researchers need to establish that predictions are reproducible, measurements are accurate and proposed explanations are consistent with experimental evidence.
Chew’s comments suggest that the next stage of AI development may be defined increasingly by the discoveries it helps scientists make, rather than solely by improvements in chatbots and digital assistants.
Why AI Could Transform Scientific Research
Scientific discovery often involves exploring enormous numbers of possibilities. Researchers may need to test different molecular structures, simulate physical systems, identify patterns in experimental data or evaluate candidate materials before finding a promising result.
AI can help reduce the time required for some of these tasks by learning patterns from existing data and generating predictions about new possibilities.
Biotechnology and drug discovery
Biotechnology is one of the most advanced areas of AI-assisted scientific research. AI systems can help predict protein structures, analyze biological interactions and design candidate molecules that may have useful properties.
Traditional drug discovery can require years of research, extensive laboratory experiments and clinical testing. AI may help researchers prioritize candidates and reduce the number of possibilities that need to be tested experimentally.
However, a computational prediction is not the same as an effective medicine. A promising molecule must still undergo laboratory studies, safety assessments and clinical trials before it can be approved for use.
Physics and complex simulations
Physics research frequently relies on mathematical models, simulations and large experimental datasets. AI can help researchers analyze observations, identify patterns and approximate calculations that would otherwise require significant computing resources.
Potential applications include materials modeling, particle physics, plasma research and the study of complex physical systems.
AI could also help scientists identify relationships that are difficult to detect through conventional analysis. Researchers would still need to determine whether those patterns represent genuine physical relationships or artifacts of the data.
Chemistry and molecular design
Chemistry involves understanding how substances interact and how molecular structures influence their properties. AI models can help predict molecular behavior, propose candidate compounds and estimate which reactions may be worth investigating.
These tools could help researchers explore chemical possibilities more efficiently, including potential catalysts, industrial chemicals and pharmaceutical compounds.
The strongest results are likely to come from combining AI predictions with established chemical knowledge and experimental verification.
Materials science
Materials science focuses on developing substances with specific characteristics, such as improved strength, lower weight, greater heat resistance or better electrical conductivity.
AI can help researchers screen candidate materials and estimate their properties before producing them in a laboratory. This could support work on batteries, semiconductors, solar cells and industrial manufacturing.
The practical value depends on whether predicted materials can be synthesized, manufactured consistently and tested under real operating conditions.
ByteDance’s Protenix-v2 Shows How AI Is Entering Science
One example of ByteDance’s scientific AI research is Protenix-v2, developed by its Seed research team. The system focuses on biomolecular structure prediction and design, with potential applications in drug discovery.
Understanding the three-dimensional structure of biological molecules can help scientists investigate how proteins interact with other molecules. That information can be useful when designing antibodies, studying diseases and identifying potential therapeutic candidates.
ByteDance’s research paper on Protenix-v2 reports improvements in antibody-antigen structure prediction compared with its previous version. It also describes experiments involving the design of antibody-like molecules and other biomolecular structures.
| Research area | Potential role of AI | Scientific requirement |
|---|---|---|
| Protein structure prediction | Estimate molecular structures from biological data | Compare predictions with experimental evidence |
| Antibody design | Suggest candidate molecules that bind to specific targets | Test binding, specificity and safety |
| Drug discovery | Prioritize compounds for further investigation | Conduct laboratory and clinical studies |
| Materials science | Predict useful material properties | Synthesize and measure candidate materials |
| Physics simulations | Analyze complex systems and experimental data | Verify results against established theory and observations |
Protenix-v2 illustrates how AI research at ByteDance is expanding beyond consumer-facing products. It does not, by itself, establish that a major scientific breakthrough is imminent, but it provides a concrete example of AI being applied to problems that researchers would otherwise investigate using conventional computational and experimental methods.
AI Is Already Contributing to Scientific Progress
Chew’s forecast builds on developments that are already underway. AI has become an increasingly important tool in biological modeling, molecular design, scientific literature analysis and computational research.
The key question is whether improvements in these tools will translate into discoveries that significantly change scientific understanding or lead to practical technologies.
A model may predict a protein structure more accurately, identify a promising chemical compound or help analyze an enormous experimental dataset. Those advances can be valuable without necessarily constituting a major scientific breakthrough.
A breakthrough generally requires a result that meaningfully changes what scientists understand or makes a previously difficult practical achievement possible. Examples might include a new material with important properties, a substantially improved approach to treating a disease or experimental evidence supporting a new scientific explanation.
AI could contribute to such outcomes by helping scientists explore more hypotheses and prioritize experiments. Yet the path from a promising prediction to a verified discovery can be long, particularly in fields where experiments are expensive or difficult to reproduce.
Why Scientific Breakthroughs Remain Difficult to Predict
Despite rapid improvements in AI capabilities, several challenges could affect the timeline Chew outlined.
First, scientific data can be incomplete, noisy or biased. A model trained on existing observations may struggle when it encounters conditions that differ substantially from its training data.
Second, scientific research involves causality as well as pattern recognition. An AI system may identify a correlation without establishing the underlying mechanism. Researchers must determine whether the proposed explanation is correct and whether it can be reproduced.
Third, laboratory experiments impose physical constraints. A molecule predicted to have desirable properties may be difficult to synthesize, unstable in practice or ineffective in a biological system.
Finally, many discoveries require collaboration among specialists. AI can help analyze information, but research teams must integrate theoretical work, experimental design, measurement, replication and peer review.
These limitations do not eliminate the possibility of rapid progress. They explain why the pace of AI development cannot be translated directly into a reliable schedule for scientific discoveries.
AI Agents Could Change How Scientists Work
The emergence of AI agents may further change scientific workflows. Unlike systems that simply answer questions, agents can potentially coordinate several tasks, use approved tools, analyze results and repeat parts of a workflow.
In a research setting, an AI agent could help search scientific literature, summarize previous experiments, generate analysis code, compare hypotheses and prepare a proposed experimental plan.
With suitable laboratory infrastructure, some systems could also coordinate automated instruments or analyze experimental results as they become available.
Such capabilities could help researchers move more quickly from a question to a testable hypothesis. However, automated research also introduces risks if systems make incorrect assumptions, mishandle data or conduct actions without adequate oversight.
Chew has separately warned that people may give AI agents excessive permissions without fully understanding the consequences. In scientific environments, this concern can extend to confidential research data, intellectual property and the control of expensive laboratory equipment.
The appropriate level of automation will depend on the task, the potential consequences of an error and the safeguards available.
Cybersecurity and Ethics Will Matter as AI Advances
Chew also highlighted cybersecurity risks associated with increasingly capable AI systems. He warned that users may not fully understand the permissions they grant to autonomous agents and that poorly controlled systems could perform unintended actions.
These concerns are relevant to scientific research because laboratories and technology companies often work with sensitive information. AI systems may have access to unpublished research, proprietary datasets, software repositories or specialized equipment.
Organizations will need to establish clear permission boundaries, monitor automated activity and ensure that consequential decisions remain subject to appropriate human review.
Scientific AI also raises questions about transparency and reproducibility. Researchers need to understand how AI-generated results were produced, which data informed the analysis and whether another team can reproduce the findings.
As AI becomes more deeply integrated into research, strong validation practices will be essential to maintaining scientific reliability.
What Could Happen Over the Next Three Years?
If AI capabilities continue improving, the period through 2029 could bring advances in how scientists formulate hypotheses, analyze data and design experiments. Some areas may move faster than others because they already have large datasets, effective simulation tools or automated testing infrastructure.
Biotechnology could benefit from improved molecular design and prediction. Materials science could use AI to identify promising candidates before laboratory synthesis. Physics and chemistry researchers could apply AI to complex simulations and experimental analysis.
The impact will depend on whether these tools can deliver measurable improvements over existing methods. A faster calculation is useful, but a major breakthrough requires a result that changes scientific knowledge or enables a meaningful new capability.
The most realistic way to interpret Chew’s forecast is as an expectation that AI will become a more important contributor to scientific discovery—not a promise that every major field will undergo a revolution within a fixed period.
The Bigger Picture
Chew’s prediction reflects a broader shift in the AI industry toward scientific and industrial applications. Technology companies are investing in models that can do more than generate text, images or videos, with increasing attention on systems that can help researchers understand biological processes, design materials and solve complex technical problems.
ByteDance’s Protenix-v2 is one example of this direction, showing how a company best known for consumer technology is expanding into scientific modeling. The larger opportunity is to combine AI’s ability to explore possibilities with the rigor of experimental science. Whether that produces a major breakthrough within two to three years remains uncertain, but the growing integration of AI into research is already changing how some scientific problems are approached.
Looking Ahead
The next few years will provide a clearer test of Chew’s forecast. Researchers will need to demonstrate that AI-assisted discoveries can be independently reproduced and translated into meaningful scientific or practical outcomes. Progress in molecular design, new materials and computational physics will be especially important areas to watch, although the timing and significance of individual breakthroughs cannot be predicted with certainty.
For ByteDance and other technology companies, the opportunity extends beyond building consumer AI products. Scientific models could become valuable tools for researchers, pharmaceutical developers, manufacturers and laboratories. The long-term impact will depend not only on model performance but also on data quality, experimental validation, safety and collaboration between AI developers and scientific institutions.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
