Coexistence of Artificial intelligence (AI) with academia: a perspective(2024.01)

2024.01.09 Editor W. M.

AI is not a new concept, with the field in academia perhaps originally starting in the mid-1950s. In the decades since, however, AI has moved from the confines of universities and research institutes to the mainstream, and permeates all areas of society. Cycles of groundbreaking research and massive technological evolution have begun to blur the lines between human and machine intelligence.  On the one hand, the possibilities for AI to improve many facets of our work and social lives are endless. On the other, AI taps into an age-old human condition: the fear of things that are unknown, or seemingly outside our control. In this column series, we highlight how AI is being used in the academic field, and discuss some of the key issues that researchers—and the wider public—face as we learn to coexist with AI. Before we begin, a question for the readership: how do you know a human has written this introduction?




Humanities and social sciences(2024.02)

2024.02.02 Editor W. M.

As well as reaching into the future, AI is helping nations rediscover their past. From the cursive kuzushiji texts of Japan to the artistic Fraktur script of Germany, there is a wealth of information and knowledge that to date has been accessible to only a few scholars. AI is now increasing accessibility by recognizing language characters in millions of documents and transcribing it into modern language that can reach a much broader audience. A specific case is kuzushiji, a cursive form of Japanese that is understood by less than 1 in 10,000 of the native Japanese population. A combination of machine and deep learning has empowered computers to first recognize the ancient kuzushiji characters and then convert them into contemporary Japanese. Such work would take hundreds of years for human scholars to complete. The utility of AI in this and similar cases is clear, and few would argue that AI was harmful in this context. However, the automated transcription of texts can pose challenges in the academic sphere. For example, what happens if the majority of original texts portray only one perspective of the society in which they were originally written? Would this lead to a biased perception of that society? Also, who retains ownership of the transcribed data, and how will the transcribed data be used? Academics have a responsibility to ensure that community leaders are fully aware of the scope and aims of their studies, as well as the final destination of the processed data.



AI技術は将来を見据えるだけでなく、各国において過去の再発見にも役立っている。日本の「くずし字」からドイツの芸術的なフラクトゥールのような字体に至るまで、これまでは一部の学者しか解読できなかった情報や知識が数多く存在する。現在、AIは文書に含まれる何百万もの文字を認識し、より多くの読者が馴染む現代語へと書き換えて読解のハードルを下げることに貢献している。具体的な例としては、「くずし字」がある。「くずし字」は、日本語を母国語とする1万人に1人以下の割合でしか理解できない、日本語の草書体である。機械学習とディープラーニングを組み合わせることで、古典的な「くずし字」をコンピュータ上でまず認識し、それを日本の現代語に変換することができるようになったが、このような作業は、学者でも膨大な年月を要するだろう。こういった事例や類似した事例でのAIの有用性は明らかであり、この文脈においては有害性を主張する人はほとんどいないだろう。しかし、AIで文書を自動的に書き換えることは、学術的な領域において課題となる可能性がある。例えば、原典の記述の大部分が、当時の社会を一部しか反映していないとしたらどうだろう。当時の社会に対する偏見につながりはしないだろうか。また、書き換えられた文書は誰に所有権があり、どのように利用されるのだろうか。学者には、研究範囲や目的、A Iで処理したデータの最終的な行き先について、コミュニティリーダーが十分理解していることを確認する責任がある。

Natural sciences (Biology)(2024.03)

2024.03.14 Editor W. M.

The Italian polymath Galileo Galilei famously put forward the concept that ‘Mathematics is the language of nature’. How fitting, then, that we should now be using mathematics in the form of AI to understand and interpret natural processes. The sheer complexity of biological processes–––from the molecular and cellular level right up to entire ecosystems––seems a tailor-made challenge for AI to solve. However, AI models need to be trained before they can be set to task. This training requires generation of vast datasets, which in turn is facilitated by advances in technology and experimental methods. Perhaps the most well-known application of AI in the biology field is the use of DeepMind’s AlphaFold suite of programs to predict the structure of proteins. Usually, this requires time- and labor-intensive wetlab experiments such as X-ray crystallography and cryo-electron microscopy. Crystallography remains the gold standard for determination of protein structure. However, AlphaFold can provide an excellent prediction for an average-sized protein in just 15-20 seconds, whereas the lab-based approaches take days. At the other end of the biological scale, AI may help us to conserve precious ecosystems and protect endangered species. For example, the non-profit organization, Imazon, developed the Previsia AI platform to help predict where deforestation of the Amazonian rainforest is most likely to occur. Previsia looks for new roads springing up in the jungle, and then correlates their appearance with multiple other geographical, demographic and environmental parameters. Combined with historical records of deforestation, this allows Previsia to score the likelihood of deforestation in a particular area. In another example, Rainforest Connection have harnessed Arbimon, an ‘acoustic AI’ platform that allows organizations and researchers to listen to the all the sounds within a given ecosystem at scale and over time. This facilitates the identification of novel species, and the tracking of their behavior and response to changing environmental conditions (including those with a negative impact, such as deforestation and poaching). If such approaches can be combined with real-time responses from local and national governments to mitigate the environmental threats, we can look to the future of the natural world with a healthy optimism.



「自然は数学の言葉を使って書かれている」、イタリアの博学者であるガリレオ・ガリレイがこの概念を提唱したのは有名な話である。そうであるならば、ばAIで数学を自然現象の理解や解釈に利用することはいかにもふさわしい。分子・細胞レベルから生態系全体に至るまで、生物学的プロセスは非常に複雑であるため、AIが解決すべき課題としてはうってつけなのではないだろうか。ただし、AIモデルにこういった課題を課すためには、事前にAIモデルをトレーニングしておく必要がある。トレーニングには膨大なデータセットの作成が必要とされるが、テクノロジーや実験手法の発展によって容易となってきた。生物学分野でもっともよく知られているAIの応用例は、おそらくDeepMind社が開発した、タンパク質の構造を予測するための一連のAlphaFoldプログラムの利用であろう。通常、タンパク質の構造決定には、X線結晶構造解析やクライオ電子顕微鏡など、時間と労力を要するウェットラボの実験が必要である。依然として、結晶学はタンパク質の構造を決定するためのゴールドスタンダードであり、通常、こういった手法では平均的な大きさのタンパク質の構造決定に数日単位の時間を要する。しかしながら、AlphaFoldであれば、わずか15〜20秒で優れた予測結果を得られるのである。他方の生態系規模の話でいうと、貴重な生態系の保全や絶滅危惧種の保護にAIが役立つ可能性がある。たとえば、非営利団体であるImazonは、アマゾン熱帯雨林の伐採が最も起こりそうな場所を予測するために、Previsia AI platformを開発した。Previsiaは、ジャングルの中の新しい道路を探索し、地理的、人口統計的、環境的といった複数のパラメーターを関連付け、森林伐採の過去の記録と組み合わせることで、特定の地域における森林伐採の可能性をスコア化することができる。また別の事例として、Rainforest Connectionは、ある生態系内のすべての音を大規模かつ長期的にわたり、組織や学者が聞くことができる音響AIプラットフォームであるArbimonを活用している。これにより新種生物の同定や、環境条件の変化(森林伐採や密猟など悪影響を及ぼすものを含む)に対する生物の行動や反応の追跡が容易となる。こういったアプローチと、環境上の脅威の軽減を目的とした、地方自治体や国によるリアルタイムの対応を組み合わせることができれば、自然界の未来を健全な楽観主義的に見据えることができるのではないだろうか。

AI and Computer Science(2024.04)

2024.04.19 Editor W. M.

The main goal of research in academic Computer Science is to design and optimize computers and the programs that they run. Studies in this field have real-world benefits including improved cyber security, better healthcare diagnostics and management, and more accurate predictive tools for forecasting financial markets and even the weather! What is the impact of AI in Computer Science? Before addressing this point, we need to take a step back and realize that AI itself is actually a subdiscipline of Computer Science. It can be thought of as one of the tools that Computer Science researchers use to improve the scope, power and efficiency of computing.

So how can AI be used to actually improve research in the field from which it was born? Although this might sound like a paradox, there are some clear examples of where this has been successful. For example, AI can help to improve software code by making it more efficient, or by detecting and fixing bugs. It can also strengthen cybersecurity by looking for vulnerabilities in software that may be susceptible to cyberattacks, and providing protection when an attack is launched. In the examples above, AI is trained on large datasets to recognize issues with pre-existing code. Increasingly, however, students and professionals are turning to AI to actually write code for them from scratch! Here, AI can be given human instructions written or spoken in regular language and convert them into code. Currently, this still requires a degree of human intervention to check the code is actually doing what the user intended. However, improvements in this use of AI could eventually make basic programming accessible to non-experts, and can free up the time of more experienced computer scientists so that they can perform more creative and strategic roles.






AI in Medicine(2024.05)

2024.05.24 Editor W. M.

Artificial intelligence (AI) is rapidly transforming the world of medicine, holding immense potential to revolutionize everything from disease diagnosis and treatment planning to drug discovery and scientific breakthroughs. By analyzing vast amounts of medical data and identifying complex patterns, AI is becoming a powerful tool for healthcare professionals, offering them crucial insights and aiding them in providing better patient outcomes.

AI holds great promise for addressing some key healthcare disparities. For example, many developing countries cannot afford state of the art DNA or RNA sequencing platforms that are required to match cancer patients to optimal treatments. However, images of cancer biopsies can be obtained by staining samples with low-cost reagents. Using many thousands of these images, the hope is that AI algorithms can be trained to identify the features of samples that are associated with specific DNA mutations. Clinicians in developing countries can then match these DNA mutations to specific cancer drugs to customize therapy for individual patients. At the recent American Association for Cancer Research annual meeting in San Diego, AI was prominently featured in many sessions. Translational researchers told us how they are using AI to integrate multiple types of ‘omic’ and clinical data to identify biomarkers that can predict which patients will respond to different drugs. Clinician-scientists explained that AI algorithms can be trained to make diagnoses, or predict treatment outcomes, from ‘looking’ at MRI scans. Companies showcased their generative AI platforms that are capable of generating novel compounds that may one day be advanced into clinical trials. Unlike the temporary darkness during the solar eclipse at the conference, the future looks very bright for AI applications in medicine!






The use of AI in streamlining academic processes(2024.06)

2024.06.19 Editor W. M.

In the previous articles, we have taken a look at the role of AI in specific academic disciplines. This final article provides a ‘helicopter view’ of how AI can be used to support processes that are common to research institutes across all disciplines.

Streamlined Recruitment: AI can automate tasks like screening resumes and scheduling interviews for research positions. It can also analyze researcher performance data and publications to identify potential candidates for new projects. Conversely, researchers who are on the job market can make use of some relatively simple AI tools to find prospective employers that match their career goals and personal skillsets.

Grant Writing Assistance and Progress Tracking: AI can analyze successful grant proposals to identify key elements and writing styles. Researchers can leverage this to generate outlines, suggest relevant literature, and even draft initial sections. Additionally, AI-powered dashboards can track progress on active grants, highlighting milestones and upcoming deadlines, ensuring timely completion and compliance.

Demand Forecasting and Inventory Management: AI can also help the purchasing staff at academic institutes in their analysis of past ordering patterns and research project timelines to predict future material needs. This allows for proactive ordering, reducing delays and preventing stockouts. AI can also optimize inventory levels, minimizing storage costs and ensuring researchers have the materials they need when they need them. Additionally, AI can identify frequently reordered items and suggest bulk purchases for cost savings.

Despite these clear benefits, it is important to acknowledge the limitations of AI prior to deploying it in key support areas.

Creativity and Originality: AI struggles with tasks requiring genuine creativity and out-of-the-box thinking. Grant proposals often hinge on novel research ideas and innovative methodologies, which AI can't replace. Human expertise remains crucial for crafting compelling narratives and justifying the significance of the proposed research.

Ethical Considerations: Bias in AI algorithms can lead to unfair hiring practices or skewed grant recommendations. Researchers need to be aware of these limitations and carefully vet AI tools to ensure they are unbiased and ethically sound. Also, when used to assist grant writing, the grant applicant(s) must ensure that they take steps to check and modify AI-generated text and tailor it to their own research prior to submission. Otherwise they run the risk of being labelled plagiarists. Ultimately, AI should be seen as a valuable tool to empower human researchers, not replace them. By leveraging the strengths of AI for administrative tasks and initial guidance, researchers can free up time and mental space to focus on the core aspects of research: conceptualizing groundbreaking ideas and conducting rigorous investigations.




採用活動の合理化: AIは、研究職の履歴書のスクリーニングや面接の日程調整といった作業を自動化することができる。また、研究者の業績データや論文を分析し、新規プロジェクトの候補者を特定することもできる。逆に、就職活動中の研究者は、比較的シンプルなAIツールを活用することで、自分のキャリア目標や個人的なスキルセットにマッチする就職希望先を見つけることができる。

助成金申請の支援と進捗管理: AIは、成功した助成金提案書を分析し、重要な要素や書き方を特定することができる。研究者はこれを活用して、アウトラインを作成し、関連文献を提案し、最初のセクションの草稿を作成することもできる。さらに、AIを活用したダッシュボードは、アクティブな助成金の進捗状況を追跡し、マイルストーンや今後の期限を強調表示することで、タイムリーな完了とコンプライアンスを保証します。

需要予測と在庫管理: AIは、学術機関の購買スタッフが過去の発注パターンと研究プロジェクトのスケジュールを分析し、将来の材料ニーズを予測する際にも役立ちます。これにより、先手を打った発注が可能になり、発注の遅れを減らし、在庫切れを防ぐことができる。また、AIは在庫レベルを最適化し、保管コストを最小限に抑え、研究者が必要な時に必要な材料を確保することができる。さらに、AIは頻繁に再注文される品目を特定し、コスト削減のための大量購入を提案することもできる。


創造性と独創性: AIは、真の創造性と既成概念にとらわれない思考を必要とするタスクに苦戦する。助成金の提案では、斬新な研究アイデアや革新的な方法論が鍵となることが多いが、これはAIでは代替できない。説得力のある物語を作り上げ、提案された研究の重要性を正当化するには、人間の専門知識が不可欠であることに変わりはない。

倫理的配慮: AIアルゴリズムのバイアスは、不公正な雇用慣行や偏った助成金推薦につながる可能性がある。研究者はこのような限界を認識し、AIツールが公平で倫理的に健全であることを確認するため、慎重に吟味する必要がある。また、助成金申請者は、助成金申請の補助に使用する場合、提出前にAIが生成した文章を確認・修正し、自身の研究に合うように調整する手順を確実に踏まなければならない。そうしないと、盗作者というレッテルを貼られる危険性があるからだ。