信任 & 合规性

数据隐私

我们维护一项产品隐私政策，解释我们的个人信息收集、使用和披露实践，以及您对个人信息的权利。我们还在下面包括了一些针对我们的教育机构客户的隐私实践的精选和一目了然的信息：

1. 地区所有权：我们并不拥有我们保管的学生个人信息。关于我们如何使用和维护学生个人信息，我们仍然直接受教育机构的控制。
2. 我们对学生个人信息的使用：我们在支持我们客户的教育目的时使用我们手头的学生个人信息。我们不会将学生个人信息用于任何广告目的，包括针对性广告目的。
3. 数据保留：我们在合同到期或终止后的90天内删除学生的个人信息。教育机构的客户可以随时要求访问、修改或删除他们学生的个人信息，我们将及时遵守这些要求。
4. 13岁以下的学生：我们遵守儿童在线隐私保护法案（COPPA）的义务。当我们的教育机构客户选择允许13岁以下的学生使用我们的产品时，我们要求教育机构确认他们有权同意我们收集13岁以下学生必要的个人信息。
5. 数据保护：我们实施并维护一个安全程序，该程序包括行业标准的行政、技术和物理保护措施，以帮助保护我们所照看的学生数据的机密性、完整性和可用性。

有关我们的隐私实践的更多信息，请查看我们的产品隐私政策。

负责任的A.I.使用

How Does Avant Use A.I.?

• From its founding in 2001, Avant’s philosophy has always been to combine the power of technology with human wisdom. We apply that same philosophy to the use of artificial intelligence (A.I.) in our products and services. We are committed to maintaining the highest standards of quality, fairness, and integrity. As part of that commitment, we have recently begun incorporating A.I. tools in select areas of our work — primarily within Mira, our subsidiary company’s language learning and formative assessment platform, within AvantProctor, to assist our human proctors by flagging potential test violations, and within our STAMP tests.

Is A.I. used in STAMP Scoring? Are all STAMP languages scored using A.I.? Is AI used in STAMP test development?

• In Avant STAMP, A.I. is currently only utilized in specific areas of test development and response scoring. These include assisting in the creation of some Reading and Listening passages, generating certain images, and scoring select responses. Currently, the automated scoring system is used only for rating a percentage of written responses in Avant STAMP 4S and 4Se Spanish, 4S French, 4S Mandarin Chinese, and 4S English. For spoken responses, 4S Spanish is currently the only language partially scored by the automated system.
• Importantly, the performance of all A.I.-based tools used in STAMP is closely monitored by our Content Development and Research teams. In the case of the scoring of select responses, sophisticated statistical analyses are conducted to assess the performance of both our human raters and the A.I. tools. These A.I. tools are only deployed when they have been rigorously validated to ensure that they uphold the same standards of quality, consistency, and accuracy as human processes. In fact, for A.I.-scored responses, our in-house machine-learning models have been shown to match – and in some cases exceed – the already high accuracy and reliability of our human raters.

Do all writing and speaking responses for those languages get rated by A.I.?

• Our system only rates responses whose proficiency level it can easily and confidently identify. Responses whose level is more difficult to determine or that may be outliers are passed to a human rater to rate.

What do you mean by “A.I.” when referring to the scoring system?

• Our system is not powered by, connected to, or integrated with third-party generative AI platforms such as ChatGPT™ or Gemini™. Avant’s automated scoring system is a machine-learning model that was developed in-house and deployed for automated scoring on Avant’s cloud-based servers. Our automated scoring system goes through rigorous testing to ensure the accuracy of its outputs. It is not connected to or influenced by unknown or unreviewed data on the internet.

Is there a chance at Avant, that an entire STAMP test can be solely rated by AI with no human rater involved?

• Because AI is not used in the scoring of Reading and Listening, there is a 0% chance of an entire STAMP test being scored by AI.
• While it is technically possible for the productive skills to both be rated by our automated system, it is highly unlikely due to the way our rating system works. For the languages for which Avant currently makes use of its proprietary, automated scoring system for writing (listed above), at least 90% – and often a considerably higher percentage –of all individual Writing sections involve rating by human raters.

How does Avant mark/identify which responses have been rated by AI?

• Our system tracks every response rated by every rater–human or automated. Our system is proprietary, and we are not at liberty to reveal whether a response was rated by a human or the automated system.

Are the scores given by A.I. checked by a human rater? If so, do the raters see the A.I. score beforehand?

• We have always ensured that at least 20% of all individual responses, whether human-scored or auto-scored, are double-scored by human raters, as is standard among testing organizations. This means that 20% of the individual responses that the automated system rates are automatically also rated by a human. If there is a disagreement between the scores assigned by the human rater and the auto-scoring tool, then a human Rater Manager determines the final score.
• Additionally, our rating quality assurance verification processes have always included regular monitoring by our research team. Should any rater, whether a human rater or the auto-scoring tool, demonstrate their rating to be any less accurate than our high standards require, they are retrained until they demonstrate the necessary improvements in accuracy.
• No rater is allowed to view another rater’s scores unless they are a Rater Manager, which ensures that ratings are independent from one another and that no one rater’s assigned proficiency level is influenced by that of another rater. This includes the automated system. If there is a discrepancy between scores given by two different raters, it automatically goes to a human Rater Manager to review and make the final rating decision.

Can you provide some data on how often a human rater may alter an A.I.-given score?

• As described above, no rater other than a Rater Manager is allowed to alter another’s score–whether human or not. See my answer above to what happens if there is a discrepancy.
• Avant prides itself on maintaining the highest rating and accuracy standards, as evidenced by our published research. Those standards continue to guide our processes, and we stand by the accuracy of both our human and automated raters.

我们将人工智能视为不是对人类专业知识的替代，而是增强和支持它的工具。通过将人类的洞察力与精心控制的人工智能相结合，我们确保我们的评估对所有的考生和依赖我们数据的组织保持有效，可靠和值得信赖。